A mine fire monitoring sensor laying method
By establishing a three-dimensional model of the flow field in mine tunnels during mine fire monitoring, extracting and processing feature vectors, and determining the optimal monitoring points, the problem of monitoring blind spots and redundancy in complex fire scenarios by sensor deployment methods is solved, and the accurate capture of the initial concentration field of a fire and the efficient reconstruction of the global concentration field are achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- WUHAN UNIV OF SCI & TECH
- Filing Date
- 2026-04-10
- Publication Date
- 2026-07-10
AI Technical Summary
Existing technologies cannot effectively capture the evolution of concentration fields in long-distance roadways in mine fire monitoring, resulting in low global concentration field inversion accuracy, making it difficult to support high-dimensional dynamic planning of evacuation paths. Furthermore, sensor deployment methods are prone to creating monitoring blind spots or redundancy in complex fire scenarios.
By establishing a three-dimensional model of the flow field in the shaft, extracting the threshold response time feature vector and the maximum time-varying gradient feature vector, performing nonlinear logarithmic mapping and singular value decomposition, constructing an augmented feature matrix, and using orthogonal triangular decomposition to determine the physical coordinates of the optimal monitoring point, sensors are then deployed.
It achieves accurate capture of the concentration field in the early stage of a fire, improves the robustness of the global concentration field reconstruction, ensures that the deployment points are highly consistent with the dynamic characteristics of fire smoke, and provides reliable data support for disaster prevention decision-making and path planning.
Smart Images
Figure CN122366254A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of mine safety monitoring technology, and more specifically to a method for deploying mine fire monitoring sensors. Background Technology
[0002] Currently, during mine fires, toxic fumes, such as CO, spread rapidly along complex tunnel networks. To effectively guide emergency rescue and disaster avoidance decisions, sensors need to be scientifically deployed within the tunnels to dynamically analyze the spatiotemporal distribution of concentration fields in real time. Currently, most mine sensor deployments follow safety regulations, primarily located in the return airflow of the mining face, major underground junctions, and densely populated areas. This traditional, experience-based deployment method only achieves isolated point-to-point monitoring and cannot effectively capture the evolution of concentration fields over long distances, resulting in low accuracy in global concentration field analysis and difficulty in supporting high-dimensional dynamic planning of evacuation paths.
[0003] To address the optimization issues of the aforementioned monitoring deployment, existing technologies mainly employ the following three representative methods, but all of them have significant drawbacks in complex fire scenarios:
[0004] The method of uniformly spaced points (such as patent CN110031086A): This type of method mechanically follows the geometric spacing, completely ignoring the non-uniformity of CO concentration distribution in the flow field, resulting in monitoring redundancy in the concentration stable area, while it is easy to form a monitoring blind zone in the high concentration gradient diffusion area (such as near the fire source).
[0005] Maximum variance methods based on POD or PCA dimensionality reduction (such as patent CN113340578A): their placement criteria are dominated by the absolute magnitude of the data. In fire scenarios, this method is easily masked by the high-concentration steady-state data at the fire source, resulting in a large number of sensors being stacked at the center of the fire source, thus completely losing the ability to capture the low-concentration spread front in the early stage of the fire.
[0006] Purely data-driven K-Means clustering methods (such as patent CN112906354A) select monitoring points based solely on the geometric clustering characteristics of the data. Due to the lack of physical constraints such as "abrupt concentration gradient" and "time of arrival of the frontier," the monitoring points selected by this method lack physical continuity and are difficult to meet the high timeliness requirements of fire early warning.
[0007] Therefore, how to provide a method that can avoid the masking of low-concentration spread front signals by high-concentration steady-state data and achieve effective monitoring of the early stages of a fire is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention
[0008] In view of the above problems, the present invention is proposed to provide a method for deploying mine fire monitoring sensors to overcome or at least partially solve the above problems.
[0009] To achieve the above objectives, the present invention adopts the following technical solution: A method for deploying mine fire monitoring sensors specifically includes the following steps: S1. Establish a three-dimensional model of the flow field in the shaft and tunnel, and use numerical simulation to obtain data on the evolution of CO concentration over time at each spatial node under a fire scenario. N Each spatial node T The concentration data at each time step are arranged to construct a dimension. N×T The original spatiotemporal concentration matrix X .
[0010] S2, from the original spatiotemporal concentration matrix X Extract the threshold response time feature vector that characterizes the dynamic features in the early stage of a fire. With the maximum time-varying gradient eigenvector F grad .
[0011] S3. Perform nonlinear logarithmic mapping and centering on the original spatiotemporal concentration matrix X to obtain the logarithmic characteristic matrix. ψ ( X ).
[0012] S4. For the logarithmic characteristic matrix ψ ( X Perform singular value decomposition according to the logarithmic characteristic matrix. ψ ( X Before extracting the singular value energy percentage in ) r Each left singular vector constructs a dimension of... N×r The spatial characteristic basis matrix Φ.
[0013] S5. Combine the spatial feature basis matrix Φ and the threshold response time feature vector. With the maximum time-varying gradient feature vector F grad Perform column-oriented feature concatenation and introduce feature weight penalty coefficients to construct a dimension of N× ( r The augmented feature matrix W is +2).
[0014] S6. Perform a transpose operation on the augmented feature matrix W to obtain the augmented transpose matrix. W T For the augmented transpose matrix W T Perform orthogonal triangular decomposition with column pivot selection to obtain permutation matrix P. Determine the physical coordinates of n optimal monitoring points based on the row indices of the non-zero elements in the first n columns of permutation matrix P.
[0015] Preferably, the threshold response time feature vector is extracted in S2. The specific method is as follows: Identify the first time the concentration in each row of the original spatiotemporal concentration matrix X reaches a preset concentration threshold. Time step index Then normalize it:
[0016] in, To prevent the minimum constant from overflowing to zero, a preset concentration threshold is used. The specific settings depend on the type of fire source.
[0017] Preferably, the maximum time-varying gradient feature vector is extracted in S2. F grad The specific method is as follows: Calculate the first-order difference of each row of the original spatiotemporal concentration matrix X in the time dimension, extract the maximum concentration abrupt change value between adjacent time steps of each spatial node, and perform normalization processing to obtain the maximum time-varying gradient feature vector. F grad .
[0018] Preferably, in S3, the original spatiotemporal concentration matrix... X Perform nonlinear logarithmic mapping and centering processing, specifically including: processing the original spatiotemporal concentration matrix. X The logarithmic feature matrix is obtained by taking the logarithm of each element and then centering it. ψ ( X ):
[0019] in, To prevent the constant from underflowing in logarithmic operations, This is the mean matrix obtained by taking the mean of the original spatiotemporal concentration matrix after logarithmic transformation.
[0020] Preferably, the feature weight penalty coefficient introduced in S5 includes the time response weight coefficient. α and concentration gradient weighting coefficient β The augmented feature matrix W The construction formula is:
[0021] in, For time response weighting coefficients, This represents the concentration gradient weighting coefficient.
[0022] Preferably, when the focus is on rapid response capability in the early stages of a fire, the time response weighting coefficient is increased. αThe value of is determined by increasing the concentration gradient weighting coefficient when the focus is on capturing the severity of concentration abrupt changes. β The value of ; in the standard reconstruction scenario, the time response weight coefficient α With the concentration gradient weighting coefficient β All values are 1.0.
[0023] Preferably, the decomposition formula for orthogonal triangular decomposition with column pivoting in S6 is:
[0024] in, Q It is an orthogonal matrix. R It is an upper triangular matrix. P Let be the permutation matrix.
[0025] Preferably, the original spatiotemporal concentration matrix in S1 X The rows correspond to spatial nodes, the columns correspond to time steps, and each element represents a spatiotemporal concentration matrix. X The CO concentration value of the corresponding spatial node at the corresponding time step.
[0026] Preferably, in S6, sensors for monitoring mine fires are deployed at the determined n optimal monitoring points.
[0027] As can be seen from the above technical solutions, compared with the prior art, the present invention discloses a method for deploying mine fire monitoring sensors. The beneficial effects of the above technical solutions provided by the embodiments of the present invention include at least the following: ① It ensures that the spatial distribution of deployment points is highly consistent with the dynamic evolution characteristics of fire smoke, and can accurately capture the spatiotemporal abrupt change patterns of the concentration field.
[0028] ② It improves the robustness of global concentration field reconstruction, providing reliable data support for disaster prevention decision-making and path planning in complex tunnel environments.
[0029] ③ It effectively balances the impact of extreme magnitude differences in the physical field on the selection criteria, realizes global collaborative capture of high-value fire source areas and low-concentration spread areas, and solves the problem that key leading-edge signals in the early stage of a fire are easily masked by background variance.
[0030] ④ A deployment logic with high physical interpretability was established to ensure that the deployment points are highly consistent with the dynamic evolution characteristics of fire smoke. Attached Figure Description
[0031] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.
[0032] Figure 1 This is a flowchart of the mine fire monitoring method provided in this embodiment of the invention; Figure 2 This is a schematic diagram illustrating the construction principle of the original spatiotemporal concentration matrix provided in this embodiment of the invention; Figure 3 This is a schematic diagram illustrating the extraction of physical early warning feature vectors provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of logarithmic flow mapping and spatial feature basis extraction provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of augmented feature matrix splicing and fusion provided in an embodiment of the present invention; Figure 6 This is an orthogonal triangular decomposition optimization and physical coordinate mapping diagram provided in the embodiments of the present invention; Figure 7 The distribution of selected locations for each monitoring method provided in the embodiments of the present invention; Figure 8 This is an error comparison diagram of the strip-shaped room-column method provided in the embodiments of the present invention; Figure 9 This is an error comparison diagram of the upward layered filling method provided in the embodiments of the present invention; Figure 10 This is an error comparison diagram of the bottomless column segmented collapse method provided in the embodiments of the present invention. Detailed Implementation
[0033] 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 some embodiments of the present invention, and not all embodiments. 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.
[0034] like Figure 1 As shown in the figure, an embodiment of the present invention discloses a method for deploying mine fire monitoring sensors, which specifically includes the following steps: S1. Establish a three-dimensional model of the flow field in the shaft and tunnel, and use numerical simulation to obtain data on the evolution of CO concentration over time at each spatial node under a fire scenario. N Each spatial node TThe concentration data at each time step are arranged to construct a dimension. N×T The original spatiotemporal concentration matrix X .
[0035] S2, from the original spatiotemporal concentration matrix X Extract the threshold response time feature vector that characterizes the dynamic features in the early stage of a fire. F time With the maximum time-varying gradient eigenvector F grad .
[0036] Threshold response time feature vector The time step index used to identify the first time the concentration of each node reaches a preset threshold (e.g., 50 ppm) is obtained after normalization and is used to characterize the lead speed of fire spread.
[0037] Maximum time-varying gradient eigenvector The maximum first-order difference of the concentration at each node in the time dimension is calculated and obtained after normalization, which is used to characterize the degree of abrupt change in fire hazard.
[0038] Among them, the threshold response time feature vector is extracted. The specific method is as follows: The concentration in each row of the identification matrix X first reaches the preset concentration threshold. Time step index Then normalize it:
[0039] in, To prevent the overflow of a very small constant from zero; a preset concentration threshold is used. The specific setting depends on the type of fire source; here, the value is 50 ppm.
[0040] Extracting the maximum time-varying gradient feature vector The specific method is as follows: Calculate the first-order difference of each row of matrix X along the time dimension, extract the maximum concentration mutation value between adjacent time steps of each node, and perform normalization to obtain the column vector. .
[0041] S3. Perform nonlinear logarithmic mapping and centering on the original spatiotemporal concentration matrix X to obtain the logarithmic characteristic matrix. ψ ( X ).
[0042] Specifically, the formula for calculating the logarithmic characteristic matrix is:
[0043] In the formula, The constant used to prevent underflow in logarithmic operations has a range of values of 1. ; This is the mean matrix obtained by taking the mean of the original spatiotemporal concentration matrix after logarithmic transformation.
[0044] S4. For the logarithmic characteristic matrix ψ ( X Perform singular value decomposition according to the logarithmic characteristic matrix. ψ ( X Before extracting the singular value energy percentage in ) r Each left singular vector constructs a dimension of... N×r The spatial characteristic basis matrix Φ.
[0045] S5. Combine the spatial feature basis matrix Φ and the threshold response time feature vector. F time With the maximum time-varying gradient feature vector F grad Perform column-oriented feature concatenation and introduce feature weight penalty coefficients to construct a dimension of N× ( r +2) Augmented characteristic matrix W .
[0046] Specifically, the augmented feature matrix W The construction formula is:
[0047] In the formula, For time response weighting coefficients, These are the concentration gradient weighting coefficients. and The range of values is .
[0048] Using the augmented characteristic matrix formula Feature splicing is performed, and the physical prior features are embedded into the mathematical feature space as mandatory constraints to adjust the ratio between physical features and mathematical variance.
[0049] The above weighting coefficients With concentration gradient weighting coefficient The value is dynamically adjusted based on the warning requirements: When the focus is on rapid response capabilities in the early stages of a fire, increase The value of is such that it satisfies ; When focusing on capturing the severity of concentration changes, increase The value of is such that it satisfies ; In standard reconstruction scenarios and All values are set to 1.0 to achieve a balance constraint between physical prior features and data-driven spatial basis matrices.
[0050] S6. Perform a transpose operation on the augmented feature matrix W to obtain the augmented transpose matrix. W T The augmented transpose matrix W T Perform orthogonal triangular decomposition with column pivot selection to obtain permutation matrix P. Based on the row indices of the non-zero elements in the first n columns of the permutation matrix P, determine the physical coordinates of n optimal monitoring points. Using the physical coordinates of the optimal monitoring points, deploy sensors for monitoring mine fires to achieve effective monitoring and early warning of the initial stage of a fire.
[0051] The QR decomposition formula with column pivoting is as follows:
[0052] In the formula, Q is an orthogonal matrix, R is an upper triangular matrix, and P is a permutation matrix with point selection priority sorting function.
[0053] The following detailed description of a mine fire monitoring sensor deployment method according to the present invention, with reference to specific embodiments and accompanying drawings, provides a detailed explanation.
[0054] This embodiment uses a 32m long mine tunnel at the fire source as the application scenario. Fluent is used to simulate the CO diffusion process after a fire. 206 monitoring nodes are extracted by discretizing the tunnel's grid space, with each node recording CO concentration data at 179 equally spaced time steps. The specific operation steps of this method are as follows: The 206 nodes were arranged along the roadway direction, and the 179 time-step concentration data contained in each node were arranged in time sequence, as follows: Figure 2 As shown, this constructs a dimension of The original spatiotemporal concentration matrix X. Each row of this matrix represents the temporal concentration curve of a specific physical location, and each column represents a snapshot of the concentration distribution of the entire tunnel at a certain moment.
[0055] like Figure 3 As shown, for the matrix Feature engineering is performed to extract two core physical vectors: the threshold response time feature vector. and the maximum time-varying gradient eigenvector .
[0056] Threshold response time feature vector Set early warning thresholds Traversing the matrix X For each row, retrieve the index of the time step when the concentration first reaches 50 ppm. If a node never reaches the threshold during the observation period, then let it... The maximum time step is 179. Normalization is performed on all index values, resulting in a dimension of... column vectors .
[0057] Maximum time-varying gradient eigenvector : For the matrix X First-order differences are calculated along the time axis to obtain the concentration change rate between adjacent time steps at each node. The maximum value in each row of the difference sequence is extracted to characterize the potentially drastic concentration abrupt change at that location. After normalization, a dimension of [dimensionality missing] is constructed. column vectors .
[0058] Considering that the concentration difference between the ignition source center and the diffusion front can reach several orders of magnitude, using the formula For matrix X Perform nonlinear mapping. This is used to avoid the risk of underflow in the zero-concentration region under logarithmic operations. Subsequently, the logarithmized matrix is centered by removing the mean from each column to eliminate steady-state components and highlight dynamic fluctuation characteristics.
[0059] Next, singular value decomposition (SVD) is performed on the centered logarithmic eigenma matrix, and the solution process is as follows: Figure 4 As shown. By truncating the first 10 principal components of the left singular matrix, a matrix with dimension... Spatial characteristic basis matrix .
[0060] Combination such as Figure 5 The splicing and fusion process shown will merge the matrix With physical eigenvectors , Perform concatenation and set feature weight penalty coefficients. α =1.0 and β =1.0, the formula is as follows This operation constructs a dimension that is expanded to... augmented feature matrix W This matrix not only preserves the variance characteristics of the global flow field, but also explicitly embeds the physical priors of concentration abrupt changes and danger arrival times.
[0061] like Figure 6 As shown, for the constructed matrix W its transpose matrix (dimension is) Perform QR decomposition with column pivoting selection, i.e., calculate .
[0062] Assume the project's preset sensor deployment number is 15. From the permutation matrix... P The first 15 principal indices are extracted and reverse-mapped to the original 206 node coordinates. These 15 coordinate points are the optimal monitoring locations determined by this invention.
[0063] To verify the effectiveness of the sampling point deployment, this embodiment uses the conformal piecewise cubic interpolation (PCHIP) algorithm to reconstruct the concentration field of all 206 nodes using only the sampling data from these 15 monitoring points.
[0064] like Figure 7 Experimental analysis shows that the method of the present invention achieves reasonable densification of monitoring points on both the intake air side (CO spread front) and the return air side of the fire source in the roadway, effectively balancing "extreme value monitoring" and "front-end early warning".
[0065] like Figure 8 As shown, the comparison of the mean absolute errors of various monitoring methods in the strip-shaped room-column method fire source tunnel scenario is presented. Figure 9 The comparison of the mean absolute errors of various monitoring methods in the fire source roadway scenario using the upward layered filling method is presented. Figure 10 This paper compares the mean absolute errors (MAEs) of various monitoring methods in a fire-prone roadway scenario using the bottomless column segmental collapse method. Specifically, in the upward hierarchical filling method scenario, the purely data-driven K-Means clustering method suffers a high error of 243.63 due to the lack of physical constraints. In contrast, the method of this invention, by embedding prior physical features such as threshold response time and maximum time-varying gradient, achieves an MAE of only 62.85. In the bottomless column segmental collapse method scenario, the performance of this invention is comparable to that of the K-Means clustering method (MAEs of 0.114 and 0.113, respectively). However, this invention is a deterministic algorithm, requiring no random initialization and ensuring repeatable results. In the strip-shaped room-column method scenario, this invention is slightly inferior to the K-Means clustering method, but the difference is not significant, and it still outperforms other methods. Based on the combined validation in three scenarios, the method of this invention demonstrates stronger robustness and physical interpretability in complex mine fire monitoring.
[0066] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.
[0067] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A method for deploying mine fire monitoring sensors, characterized in that, Specifically, the following steps are included: S1. Establish a three-dimensional model of the flow field in the shaft and tunnel, and use numerical simulation to obtain data on the evolution of CO concentration over time at each spatial node under a fire scenario. N Each spatial node T The concentration data at each time step are arranged to construct a dimension. N×T The original spatiotemporal concentration matrix X ; S2, from the original spatiotemporal concentration matrix X Extract the threshold response time feature vector that characterizes the dynamic features in the early stage of a fire. With the maximum time-varying gradient eigenvector F grad ; S3. Perform nonlinear logarithmic mapping and centering on the original spatiotemporal concentration matrix X to obtain the logarithmic characteristic matrix. ψ ( X ); S4. For the logarithmic characteristic matrix ψ ( X Perform singular value decomposition according to the logarithmic characteristic matrix. ψ ( X Before extracting the singular value energy percentage in ) r Each left singular vector constructs a dimension of... N×r The spatial characteristic basis matrix Φ; S5. Combine the spatial feature basis matrix Φ and the threshold response time feature vector. With the maximum time-varying gradient feature vector F grad Perform column-oriented feature concatenation and introduce feature weight penalty coefficients to construct a dimension of N× ( r The augmented characteristic matrix W of +2); S6. Perform a transpose operation on the augmented feature matrix W to obtain the augmented transpose matrix. W T For the augmented transpose matrix W T Perform orthogonal triangular decomposition with column pivot selection to obtain permutation matrix P. Determine the physical coordinates of n optimal monitoring points based on the row indices of the non-zero elements in the first n columns of permutation matrix P.
2. The method for deploying mine fire monitoring sensors according to claim 1, characterized in that, Extracting the threshold response time feature vector from S2 The specific method is as follows: Identify the first time the concentration in each row of the original spatiotemporal concentration matrix X reaches a preset concentration threshold. Time step index Normalization is performed: in, To prevent the minimum constant from overflowing to zero, a preset concentration threshold is used. The specific settings depend on the type of fire source.
3. The method for deploying mine fire monitoring sensors according to claim 1, characterized in that, Extracting the maximum time-varying gradient feature vector from S2 F grad The specific method is as follows: Calculate the first-order difference of each row of the original spatiotemporal concentration matrix X in the time dimension, extract the maximum concentration abrupt change value between adjacent time steps of each spatial node, and perform normalization processing to obtain the maximum time-varying gradient feature vector. F grad .
4. The method for deploying mine fire monitoring sensors according to claim 1, characterized in that, S3 on the original spatiotemporal concentration matrix X The logarithmic characteristic matrix is obtained by performing nonlinear logarithmic mapping and centering processes. ψ ( X ): in, To prevent constants from underflowing in logarithmic operations, This is the mean matrix obtained by taking the mean of the original spatiotemporal concentration matrix after logarithmic transformation.
5. The method for deploying mine fire monitoring sensors according to claim 1, characterized in that, The feature weight penalty coefficient introduced in S5 includes the time response weight coefficient. α and concentration gradient weighting coefficient β The augmented feature matrix W The construction formula is: in, For time response weighting coefficients, This represents the concentration gradient weighting coefficient.
6. The method for deploying mine fire monitoring sensors according to claim 5, characterized in that, When the focus is on rapid response capabilities in the early stages of a fire, the time response weighting coefficient should be increased. α The value of is determined by increasing the concentration gradient weighting coefficient when the focus is on capturing the severity of concentration abrupt changes. β The value of ; in the standard reconstruction scenario, the time response weight coefficient α With the concentration gradient weighting coefficient β All values are 1.
0.
7. The method for deploying mine fire monitoring sensors according to claim 1, characterized in that, The decomposition formula for orthogonal triangular decomposition with column pivoting in S6 is: in, Q It is an orthogonal matrix. R It is an upper triangular matrix. P Let be the permutation matrix.
8. The method for deploying mine fire monitoring sensors according to claim 1, characterized in that, Original spatiotemporal concentration matrix in S1 X The rows correspond to spatial nodes, the columns correspond to time steps, and each element represents a spatiotemporal concentration matrix. X The CO concentration value of the corresponding spatial node at the corresponding time step.
9. The method for deploying mine fire monitoring sensors according to claim 1, characterized in that, In S6, sensors for monitoring mine fires are deployed at the n optimal monitoring points.