A harmful gas flow direction identification and aggregation direction prediction method based on multi-sensor fusion

Abstract

The present application provides a kind of based on multi-sensor fusion harmful gas flow direction identification and aggregation direction prediction method, it is related to mine tunnel prevention and control technical field, the present application obtains the real wind power and gas information in each point of mine tunnel by wind direction indicator and gas monitor, and based on the mine tunnel form, constructs the mine tunnel three-dimensional model and virtual point position containing same coordinate system.The real wind power facility and known harmful gas release point are mapped as input condition and first escape condition, while randomly adding second escape condition for fluid simulation, the similarity of virtual and real information matrix is maximized to screen the optimal scenario.Finally, set sliding time window, use the optimal scenario sequence in window as power source to drive future simulation in cycle, and update with measured data, realize the dynamic prediction of harmful gas aggregation position and concentration.

Description

Technical Field

[0001] This invention relates to the field of mine gas safety control technology, specifically a method for identifying the flow direction and predicting the accumulation direction of harmful gases based on multi-sensor fusion. Background Technology

[0002] In the mining environment, the accumulation and diffusion of harmful gases (such as methane, carbon monoxide, and hydrogen sulfide) are major causes of serious safety accidents such as explosions and poisoning. Traditional gas monitoring methods mostly use fixed sensors to detect concentrations at specific points, issuing alarms when the gas concentration exceeds a threshold. However, this method can only indicate that "the concentration has exceeded the standard at a certain location," and cannot answer dynamic questions such as "where the gas comes from, where it goes, and where it will accumulate," nor can it provide early warnings before the gas diffuses to the sensor location. With the increasing complexity of mine ventilation systems and the continuous advancement of mining operations, the interaction between the airflow field and the gas release source exhibits strong spatiotemporal variability, and static threshold alarms can no longer meet the needs of proactive safety control.

[0003] In recent years, computational fluid dynamics (CFD) simulation technology has been gradually introduced into mine gas diffusion simulation. However, relying solely on simulation has the problem of accurately defining initial boundary conditions—unpredictable factors such as unknown fissures and seepage in real mines, and release sources generated by temporary operations, often cause simulation results to deviate from actual measurements. Meanwhile, the development of IoT sensor technology has made it possible to deploy wind direction indicators and gas monitors on a large scale on the inner walls of mines, providing the hardware foundation for obtaining high-density measured wind and concentration field data. Based on this, combining real-time measurement data from multiple sensors with dynamic inversion through fluid simulation to establish a fusion method capable of continuously absorbing measured information, automatically identifying unknown release sources, and continuously predicting gas accumulation directions has become a technological development direction for improving the intelligent level of mine safety monitoring.

[0004] The existing technology, disclosed in CN114841031A, describes a method for calculating the concentration of toxic and harmful gas diffusion in a three-dimensional virtual training environment. The method comprises: S1, determining the effective influencing factors of toxic gas diffusion; S2, constructing a physical model of the corresponding training environment and deriving the effective influencing factors of gas diffusion; S3, dividing the grid into dense and sparse sections based on the gas toxicity dose and the size of the obstruction, and distinguishing grid densities; S4, constructing a CFD model to calculate the simulated concentration of harmful gas diffusion; S5, establishing an offline database of time-series concentration trend results over a period of time to store the gas concentration trend results; and S6, rapidly displaying the results in the virtual environment during training.

[0005] While the publicly available technical documents have achieved the prediction of harmful gas concentrations, they have not taken into account the changing environment inside the mine and the complex on-site conditions, making it difficult for the prediction results to reflect the actual situation.

[0006] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0007] The purpose of this invention is to provide a method for identifying the flow direction and predicting the accumulation direction of harmful gases based on multi-sensor fusion, so as to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention provides the following technical solution: A method for identifying the flow direction and predicting the accumulation direction of harmful gases based on multi-sensor fusion, comprising the following steps: Step 1: Based on the morphological data of the mine, establish a three-dimensional model of the mine, obtain the real wind force information and real gas information of each monitoring point in the mine, mark the points corresponding to each monitoring point in the three-dimensional model of the mine to form virtual points, and associate the real wind force information and real gas information with the corresponding virtual points; Step 2: Obtain the location and air source information of the ventilation facilities in the mine, and map the location and air source information to the 3D model of the mine as input conditions. Obtain the release points and release amounts of each known harmful gas in the mine, and map the release points and release amounts to the 3D model of the mine as the first escape condition of the harmful gas. Step 3: Set multiple second escape conditions in the 3D model of the mine, and perform fluid simulation based on the input conditions, the first escape conditions and each second gas release condition to obtain multiple candidate simulation scenarios, and obtain the virtual wind force information and virtual gas information of each virtual point in each candidate simulation scenario; Step 4: Match the virtual wind and gas information in each candidate simulation scenario with the real wind and gas information respectively, and determine the candidate simulation scenario with the highest matching degree as the optimal scenario; Step 5: Continuously perform simulations and obtain real wind and gas information. Set a time window and the optimal scene sequence corresponding to each stage within the time window. Use the fluid dynamics source of the optimal scene sequence to cyclically obtain the location and concentration of harmful gases in the future.

[0009] Furthermore, wind direction indicators and gas monitors are installed on the inner wall of the mine. Each wind direction indicator is equipped with a gas monitor. The wind direction indicator acquires the real wind force information at each location, which includes wind direction and wind force. The gas monitor acquires the real gas information at each location, which includes the content of various gases. An orthogonal coordinate system is established based on the mine, and the specific coordinates of each wind direction indicator in the orthogonal coordinate system are recorded.

[0010] Furthermore, the morphological data of the inner wall of the mine is scanned, and the morphological data is input into the simulation software to construct a three-dimensional model of the mine. The same coordinate system as the real mine is established in the three-dimensional model of the mine. At the same time, the wind direction indicator points that are the same as those in the real mine are marked in the three-dimensional model of the mine and marked as virtual points. Virtual wind force information and virtual gas information are obtained respectively.

[0011] Furthermore, the coordinates of each wind power facility in the mine and the wind source information it generates are obtained. The wind source information includes wind force, wind direction, and wind volume. The wind power facilities and their wind source information are marked in the three-dimensional model of the mine according to the coordinates and set as input conditions. The release points and release amounts of harmful gases in a real mine are obtained. The release amounts are obtained through on-site measuring instruments. At the same time, the release points and release amounts are marked in the three-dimensional model of the mine, and the points are marked as the first escape conditions.

[0012] Furthermore, fluid simulation is performed on the three-dimensional model of the mine shaft in simulation software. During the simulation process, a second escape condition is set, and the logic of the second escape condition is as follows: At positions other than the first escape condition, release points and release amounts are set. The release amount at each release point of the second escape condition is less than the minimum value of the first escape condition. Several candidate simulation scenarios are constructed. The logic for setting release points and release amounts for the candidate simulation scenarios is as follows: A grid is defined on the surface of the 3D model of the mine. Several grid intersections are selected as escape points and escape amounts are set to form a candidate simulation scene. Several candidate simulation scenes are formed by selecting escape points and their escape amounts multiple times, and all candidate simulation scenes are made to have no completely identical escape points and escape amounts. The pressure information inside and at the entrance of a real mine is obtained, and the fluid power source is simulated by combining the input conditions, the first escape condition and the second escape condition.

[0013] Furthermore, select the optimal scenario, including: The simulation process acquires virtual wind and gas information at each virtual point, forms a virtual information matrix based on the points, and forms a real information matrix by combining the real wind and gas information at each point in the real mine. The similarity between each virtual information matrix and the real information matrix is ​​calculated, and the simulation process corresponding to the maximum value is selected as the optimal scenario.

[0014] Furthermore, real wind and gas information are continuously acquired, and time windows are set to include the latest acquired real wind and gas information. The optimal scenario is selected for each time window to form an optimal scenario sequence. The input conditions, first escape condition, and second escape condition corresponding to each simulation process in the optimal scenario sequence are used as fluid power sources and cycled to form the overall scenario of this time window; The system performs future time simulations on the overall scene to obtain the predicted locations of harmful gas accumulation and their corresponding concentration values.

[0015] Furthermore, as the time window is updated with real wind and gas information, the overall scene and the location of harmful gas accumulation are updated simultaneously.

[0016] Compared with the prior art, the beneficial effects of the present invention are: This invention solves the problem of rigid spatial registration of multi-source heterogeneous data by constructing a 3D model of the mine with geometry consistent with the physical mine and monitoring points in the same coordinate system, transforming real wind and gas information into a truth matrix that can be directly compared point by point. By distinguishing between input conditions and the first escape condition, and introducing a constrained second escape condition, the fluid simulation is extended from deterministic calculation to a candidate pool that can invert unknown disturbances. Then, through maximizing the similarity between the virtual matrix and the real matrix, the unobservable leakage distribution is inverted into the optimal estimate. Furthermore, a sliding time window is adopted, using the optimal scene sequence at each moment within the window as a fluid dynamic source to cyclically drive future simulations, enabling the prediction model to be updated continuously with measured data, continuously calibrating the assumptions of unknown sources and accumulation locations, and ultimately achieving dynamic real-time prediction of the flow direction and accumulation direction of harmful gases, transforming the static digital twin into an online adaptive early warning system. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall method flow of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0019] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0020] Example: Please see Figure 1 The present invention provides a technical solution: A method for identifying the flow direction and predicting the accumulation direction of harmful gases based on multi-sensor fusion, comprising the following steps: Step 1: Establish a 3D model of the mine based on the morphological data of the mine, obtain the real wind force information and real gas information of each monitoring point in the mine, mark the points corresponding to each monitoring point in the 3D model of the mine to form virtual points, and associate the real wind force information and real gas information with the corresponding virtual points; Step 1 includes the following: Step 101: Install wind direction indicators and gas monitors on the inner wall of the mine. Each wind direction indicator is equipped with a gas monitor. The wind direction indicator acquires the real wind force information at each location, which includes wind direction and wind force. The gas monitor acquires the real gas information at each location, which includes the content of various gases. Establish an orthogonal coordinate system based on the mine and record the specific coordinates of each wind direction indicator in the orthogonal coordinate system.

[0021] By synchronously binding gas monitors to each wind direction indicator location on the mine's inner wall and recording the three-dimensional coordinates of each point based on a unified orthogonal coordinate system, rigid spatial registration of the three physical quantities—wind direction, wind force, and gas concentration—is achieved at the original measurement level. Point binding eliminates the risk of spatial misalignment between the wind field and concentration field caused by the separation of sensor installation locations, avoiding positioning errors introduced by coordinate interpolation or nearest neighbor matching in subsequent analysis. The introduction of the orthogonal coordinate system discretizes the irregular continuous space of the mine into a series of observation points with definite algebraic coordinates. The wind force vector (direction and magnitude) and gas concentration vector (content of each component) at each observation point together constitute a state feature element of more than five dimensions. This transforms the continuous, simulated, and difficult-to-model flow and concentration fields in the real physical world into a measurement sequence that can be repeatedly read and numerically compared on known discrete coordinates. Each coordinate point thus becomes an independent "truth anchor point," providing an unambiguous benchmark for point-by-point comparison of subsequent simulation results with the real situation. The acquisition method naturally endows the measurement data with a matrix-like structure—wind information from all points can be arranged into a wind matrix, and gas information into a gas matrix, with the row and column indices of both matrices determined entirely by the coordinate system without additional alignment. This provides a directly usable data format for constructing the "real information matrix" and calculating its similarity to the virtual information matrix in step 4, avoiding time-consuming resampling or coordinate transformations during data preprocessing. It solves the most fundamental and critical problem in multi-sensor fusion: how to accurately anchor the measurement results of different physical quantities within the same spatial reference system, thus enabling the numerical verification of the "wind-driven aerodynamics" physical process. Step 102: Scan the morphological data of the inner wall of the mine and input the morphological data into the simulation software to construct a three-dimensional model of the mine. Establish the same coordinate system as the real mine in the three-dimensional model of the mine. At the same time, mark the same wind direction indicator points as in the real mine in the three-dimensional model of the mine and mark them as virtual points. Obtain virtual wind force information and virtual gas information respectively.

[0022] As a preferred embodiment, the simulation software can be a general-purpose computational fluid dynamics (CFD) software, such as ANSYS Fluent / CFX, Siemens Star-CCM+, etc. Such software can import the geometric model of the mine surface obtained by scanning (such as STL, point cloud to surface), establish a coordinate system consistent with the real mine, set momentum source terms (fans), pressure boundaries, component release sources, and obtain the velocity field (wind information) and concentration field (gas information) by solving the Navier-Stokes equations and component transport equations. It also supports the batch generation of multiple candidate simulation scenarios by script or parameterization.

[0023] By constructing a 3D model of the mine through scanning the morphological data of the mine's inner walls, and establishing an orthogonal coordinate system identical to that of the real mine, while precisely mapping real monitoring points to virtual points, the physical mine environment, which was originally unrepeatable and subject to arbitrary intervention, is transformed into a geometrically consistent, well-defined data model that can be repeatedly calculated under any input conditions. This transformation solves the fundamental limitation of not being able to make "hypothetical adjustments" to wind power facilities or release sources in the real environment—in a physical mine, changing fan parameters or adding an unknown leak point would bring real safety risks and be costly, while in the 3D mine model, such operations can be performed an unlimited number of times at low cost. More importantly, the establishment of the same coordinate system and virtual points at the same location ensures a direct, one-to-one correspondence between the simulated wind force and gas concentration values ​​and the actual sensor measurements, eliminating the need for post-processing steps that easily introduce errors, such as coordinate transformation, nearest neighbor search, or spatial interpolation. This means that the values ​​at each virtual point output by the subsequent fluid simulation can be precisely mapped to the measured values ​​at the same physical location in the real environment. This transforms the abstract judgment of "whether the simulation is reliable" into a rigorously quantifiable mathematical problem: "the sum of the distances between the simulation vector and the real vector on N fixed coordinates." Furthermore, although the virtual points, as pre-positioned observation windows, do not yet carry specific simulation values ​​at step 102, their very existence defines the set of spatial locations that the simulation results must output. This provides a definite matrix size and row / column index order for constructing the "virtual information matrix" in step 4, allowing multiple sets of simulation results to be naturally arranged into isomorphic matrices for efficient similarity calculation. This completes a one-way lossless mapping from the physical world to the digital world, preserving the boundary constraints of the complex geometry of the mine on airflow motion while establishing a spatial reference framework for all subsequent simulations. It serves as an indispensable bridge connecting real measurements and numerical simulations.

[0024] Step 2: Obtain the location and air source information of the ventilation facilities in the mine, and map the location and air source information to the 3D model of the mine as input conditions. Obtain the release points and release amounts of each known harmful gas in the mine, and map the release points and release amounts to the 3D model of the mine as the first escape condition of the harmful gas. Step 2 includes the following: The coordinates of each wind power facility in the mine and the wind source information it generates are obtained. The wind source information includes wind force, wind direction and wind volume. The wind power facilities and their wind source information are marked in the three-dimensional model of the mine according to the coordinates and set as input conditions. The release points and release amounts of harmful gases in a real mine are obtained. The release amounts are obtained through on-site measuring instruments. At the same time, the release points and release amounts are marked in the three-dimensional model of the mine, and the points are marked as the first escape conditions.

[0025] In a real mine, the coordinates, wind direction, and air volume of each wind turbine, along with the coordinates and release amount of each hazardous gas release point, are mapped to the 3D model of the mine. These coordinates are clearly distinguished as "input conditions" and "first escape conditions," ensuring a rigorous transformation from physical engineering parameters to mathematical calculation conditions for the dynamic and material source boundaries in fluid simulation. This deconstructs two fundamentally different roles in the complex process of "wind-driven aerodynamics": input conditions define the energy source and directionality driving the airflow (e.g., forced convection generated by a fan), while the first escape condition defines the spatial entry point and intensity of hazardous gases into the flow field (e.g., known leak points or release sources). These two are not interchangeable in the simulation model—input conditions change the source terms in the momentum equation, while the first escape condition changes the source terms in the component transport equation. By labeling and storing the two types of conditions separately, subsequent simulations can freely adjust the input conditions while keeping the first escape condition unchanged. Simultaneously, each condition carries precise coordinate information, ensuring that the blower's thrust is applied to the fluid at the correct location and that harmful gases are released into the mine space from the correct wall location. This avoids spatial logical contradictions between the flow field and concentration field caused by positional deviations (e.g., wind blowing from behind the blower, or gas being released from a closed rock wall where leakage is impossible). The input conditions output in step 2, along with the first escape condition, constitute the deterministic basis for the fluid simulation in step 3—all subsequently introduced second escape conditions (random release points) will use these two known conditions as references, and their release amounts will be constrained to be less than the minimum value of the first escape condition, thus ensuring that random disturbances do not overwhelm the influence of known dominant sources. By completely transforming measurable or identifiable information from actual engineering (blower parameters, known leak points) into the "prior knowledge" of the simulation model, deviations between simulation and real data can be attributed to unknown random release sources, rather than errors in the setting of known conditions, when selecting the optimal scenario.

[0026] Step 3: Set multiple second escape conditions in the 3D model of the mine, and perform fluid simulation based on the input conditions, the first escape conditions and each second gas release condition to obtain multiple candidate simulation scenarios, and obtain the virtual wind force information and virtual gas information of each virtual point in each candidate simulation scenario; Step 3 includes the following: At positions other than the first escape condition, release points and release amounts are set. The release amount at each release point of the second escape condition is less than the minimum value of the first escape condition. Several candidate simulation scenarios are constructed. The logic for setting release points and release amounts for the candidate simulation scenarios is as follows: A grid is defined on the surface of the 3D model of the mine. Several grid intersections are selected as escape points and escape amounts are set to form a candidate simulation scene. Several candidate simulation scenes are formed by selecting escape points and their escape amounts multiple times, and all candidate simulation scenes are made to have no completely identical escape points and escape amounts. The pressure information inside and at the entrance of a real mine is obtained, and the fluid power source is simulated by combining the input conditions, the first escape condition and the second escape condition.

[0027] In a preferred embodiment, a triangular surface mesh is generated on the inner wall of the 3D model of the mine shaft, with mesh nodes naturally covering all potential escape surfaces of the mine shaft. Then, a geometric eigenvalue is calculated for each mesh node, representing the average area of ​​the triangular faces connected to that node. This area value reflects the density of the wall mesh near that location and indirectly corresponds to the relative size of the wall region at that location. After the calculation is completed, all mesh nodes are sorted from largest to smallest according to their average face area, and a fixed number is set, for example, selecting the top 10 nodes with the largest areas as escape points for this candidate simulation scenario.

[0028] The escape amount at each selected escape point is directly set to be proportional to the average patch area of ​​that node. A global scaling factor is introduced to ensure that the maximum escape amount across all selected escape points is strictly less than the minimum release amount in the first escape condition. For example, if the minimum release amount in the first escape condition is Q, the ratio of the maximum area of ​​all selected nodes to Q can be calculated first. A factor slightly smaller than this ratio is then taken, and the area of ​​each node is multiplied by this factor to obtain the escape amount at that point. In this way, larger grid nodes correspond to larger escape amounts, and smaller nodes correspond to smaller escape amounts, ensuring that all escape amounts are always constrained within the secondary dominant range.

[0029] When multiple different candidate simulation scenarios need to be generated, differentiation can be achieved by changing the number of selected points, changing the geometric features used for sorting, or adjusting the emission ratio coefficient. For example, the first scenario selects the 10 nodes with the largest area and sets the emission ratio proportionally; the second scenario selects the 15 nodes with the largest area and uses a slightly smaller ratio coefficient; the third scenario selects all nodes with an average area greater than a certain threshold, and so on. As long as at least one of the point selection set or emission assignment rules is different, it can be ensured that the generated candidate scenarios are not completely identical, and the entire process is completely repeatable and traceable, achieving the non-random scenario construction required by the scheme.

[0030] By setting release points and release rates at locations other than the first escape condition, with the release rate at each point strictly less than the minimum release rate under the first escape condition, and incorporating pressure information from the actual mine interior and entrance into the fluid dynamics source, the simulation model gains the ability to simulate "unknown or hidden leakage sources," while ensuring that such random disturbances do not overshadow the influence of known dominant sources. In actual mine environments, the monitored values ​​of harmful gases are often the result of multiple sources working together—including known fixed release points (such as exhaust outlets), background transport driven by known ventilation facilities, and unpredictable sources such as unknown fissure seepage and pipe micro-leakage. Without introducing random release sources, any anomalies in monitored values ​​that cannot be explained by known conditions would be attributed to simulation errors or measurement noise, leading to a fundamental lack of the model's ability to represent real physical processes. The introduction of random release points transforms the unobservable uncertainty of "unknown sources" into simulation variables with controllable statistical characteristics, allowing the model to calculate under various random assumptions and, through similarity screening in subsequent step 4, to inversely deduce the most probable unknown release distribution. The upper limit of the release rate is constrained to be less than the minimum value of the first escape condition, ensuring that the unknown source always remains secondary dominant, preventing the simulation results from diverging or the contribution of the known dominant source from being overwhelmed. Simultaneously, the pressure information inside and at the entrance of the actual mine serves as input to the fluid dynamics source, completing the integrity of the driving mechanism—the forced convection of the fan is not the only driving force for airflow; the pressure difference between the mine interior and the external atmosphere (e.g., due to temperature differences, depth variations, or the overall resistance distribution of the ventilation system) also generates significant natural convection and even counterflow phenomena. Ignoring pressure information would cause the simulated wind field to only reflect the pushing effect of the fan, failing to capture the backflow or stagnation regions driven by pressure difference, thus leading to systematic deviations in the prediction of gas accumulation locations. Using pressure information and input conditions (fan parameters) together as the fluid dynamics source means that the source term in the momentum equation simultaneously includes both the active source (fan) and the passive source (pressure difference), and the resulting composite wind field more closely approximates the airflow state in a real mine. This step outputs multiple sets of simulation results, each corresponding to a different set of random release hypotheses. This provides a "candidate pool" for step 4—similarity comparison is only meaningful if there are sufficiently diverse and controlled random perturbations; otherwise, if all simulations have identical input conditions, the optimal solution cannot be selected. This simulates a typical real-world scenario where "known conditions determine large-scale transport structures, while unknown perturbations determine local concentration anomalies," enabling the model to handle uncertainty within a deterministic framework.

[0031] Step 4: Match the virtual wind and gas information in each candidate simulation scenario with the real wind and gas information respectively, and determine the candidate simulation scenario with the highest matching degree as the optimal scenario; Step 4 includes the following: Select the optimal scenario, including: The simulation process acquires virtual wind and gas information at each virtual point, forms a virtual information matrix based on the points, and forms a real information matrix by combining the real wind and gas information at each point in the real mine. The similarity between each virtual information matrix and the real information matrix is ​​calculated, and the simulation process corresponding to the maximum value is selected as the optimal scenario.

[0032] By arranging the virtual wind and gas information of all virtual points in each simulation process in a fixed coordinate order to construct a virtual information matrix, and simultaneously constructing the actual information matrix from the measured values ​​of the corresponding points in the real mine, the similarity between the two matrices is calculated. Finally, the simulation process corresponding to the maximum similarity is selected as the optimal scenario. This transforms the comprehensive judgment problem of "which random release hypothesis best explains the current measured data"—a problem involving multiple variables, multiple points, and multiple physical quantities—into a numerically solvable problem of minimizing the distance between matrices. The essence of the matrix form lies in preserving spatial structure information—the similarity between the two matrices not only reflects the degree of closeness of the numerical values ​​of wind or gas concentration at each independent point, but also implicitly penalizes the distortion of spatial distribution patterns. For example, if one simulation is too high at point A and too low at point B, but another simulation is too high by the same amount at both points A and B, the former may disrupt the direction of the spatial gradient while the latter may maintain it. Matrix similarity calculation (such as cosine similarity or negative correlation based on Euclidean distance) can automatically capture this difference. Selecting the simulation process corresponding to the maximum similarity is equivalent to finding the configuration that minimizes the deviation between the simulation output and the actual observation among all candidate unknown release source configurations (different random implementations of the second escape condition). This operation inverts the unobservable second escape condition from a random variable into a definite optimal estimate—although the actual unknown leakage points in the real mine may not completely coincide with the random points set in the simulation, statistically speaking, the random release distribution corresponding to the selected optimal simulation is the best approximation of the real unknown source, and the overall deviation of its wind field and concentration field across all monitoring points is controlled within the minimum range. The "optimal scenario" output in step 4 is not an abstract identifier, but carries complete boundary condition information: the input conditions corresponding to the scenario, the first escape condition, and the optimized set of second escape conditions (the location and release amount of the random release point) together constitute the most reliable digital reproduction of the real state of the mine at that moment. This result provides a single-frame baseline for the time window operation in step 501—when it is necessary to construct the overall scene within a time window, each moment within the window requires its own optimal scene, which is obtained by frame-by-frame filtering in step 4.

[0033] Step 5: Continuously perform simulations and obtain real wind and gas information. Set a time window and the optimal scene sequence corresponding to each stage within the time window. Use the fluid dynamics source of the optimal scene sequence to cyclically obtain the location and concentration of harmful gases in the future.

[0034] Step 5 includes the following: Step 501: Continuously acquire real wind force information and real gas information, and set a time window to include the latest acquired real wind force information and real gas information. Select the corresponding optimal scenario for each time window to form an optimal scenario sequence. In a preferred embodiment, the time window can be set as a fixed-length sliding interval, the length of which is determined based on the typical response cycle of the mine ventilation system, for example, 10 minutes per window. As real wind and gas information is continuously collected, the window always uses the latest sampling time as its right endpoint, and pushes the left endpoint back 10 minutes, ensuring that the window always contains continuous measured data from the most recent 10 minutes. Each time new data arrives, the window slides to the right once, discarding the oldest data points that exceed the time range. In this way, each set of real data within the window corresponds to an optimal scenario selected in the previous steps. Arranging these optimal scenarios chronologically forms the optimal scenario sequence for subsequent iterative driving. This fixed window length setting effectively filters out the interference of instantaneous sensor noise while fully preserving the dynamic trends of recent wind field and release source changes, providing a stable and timely initial state for future time predictions.

[0035] The input conditions, first escape condition, and second escape condition corresponding to each simulation process in the optimal scenario sequence are used as fluid power sources and cycled to form the overall scenario of this time window; The system performs future time simulations on the overall scene to obtain the predicted locations of harmful gas accumulation and their corresponding concentration values.

[0036] By continuously acquiring real wind and gas information, a time window containing the latest data is set. For each stage within the window, the corresponding optimal scenario is selected to form an optimal scenario sequence. Then, the input conditions, first escape condition, and second escape condition corresponding to each simulation process in this sequence are used as a fluid dynamic source for cyclical driving, constructing the overall scenario for this time window. Finally, based on this overall scenario, future time simulations are performed to obtain the predicted location and concentration of harmful gases. This transforms the originally static, isolated single-frame optimal estimation into a dynamic extrapolation model that reflects the evolution of wind and concentration fields over time. The introduction of the time window solves the fundamental problem of "which historical data should be used as the basis for simulation"—if only the data from the latest moment is used for optimal scenario selection, the model becomes overly sensitive to external disturbances (such as instantaneous wind speed fluctuations or sensor noise), leading to drastic changes in prediction results; if all historical data is used, the correlation between early information and the current state weakens, slowing down the model's response speed to real changes. The time window strikes a perfect balance between these two aspects: it smoothly preserves recent memory, ensuring that the optimal scenario at each moment is determined based on a continuous period of observations before and after that moment, thus naturally filtering out high-frequency noise while retaining mesoscale trend information. The physical significance of cyclically driving the power sources in the optimal scenario sequence lies in reproducing the true trajectory of wind field and release source changes within the time window—ventilators in a mine may operate at different speeds at different times, the emissions from known release points may fluctuate over time, and the second escape condition (random unknown source) may also change due to crack expansion or blockage. By sequentially applying the three sets of power sources corresponding to each scenario in the sequence to the simulation, it's equivalent to having the digital twin reenact the entire wind-driven aerodynamic process along the historically observed optimal path, thus ensuring that the initial state extrapolated in the future is not a "slice" of an arbitrary moment, but rather a continuous endpoint of the entire recent evolutionary history. This approach of "driving the cycle based on the historical best sequence" is fundamentally different from steady-state extrapolation using only the optimal scenario at the last moment: the latter ignores the time-varying rate of the power source. For example, if the air volume has been continuously increasing over the past ten minutes, steady-state extrapolation will assume that the air volume remains at the latest value. However, the cycle-driven approach allows the model to experience the air volume increase from low to high in the historical phases of the simulation. By the time it reaches the current moment, the flow field already includes inertial effects and cumulative transport results, making future predictions closer to the real physical process. The final output of the location and concentration of harmful gases is no longer a single deterministic result based on fixed boundary conditions, but a comprehensive prediction that integrates all known and inverted unknown source dynamic changes in recent history. This prediction has direct operability for emergency decision-making—knowing which coordinate region the concentration peak will appear in and at what level allows for advance planning of personnel evacuation routes, deployment of emergency ventilation, or setting up of containment barriers.From the perspective of data flow, step 501 receives the optimal scene sequence arranged in chronological order output from step 4 and transforms it into a time-continuous simulation, providing the first round of prediction results for the rolling update in step 502.

[0037] Step 502: As the time window is updated with real wind and gas information, the overall scene and the location of harmful gas accumulation are updated simultaneously.

[0038] By continuously updating the time window with the latest real wind and gas information, and synchronously updating the overall scenario constructed in step 501 as well as the predicted locations and concentrations of harmful gases, the entire prediction system transforms from a one-time offline calculation into an online adaptive closed loop synchronized with the physical mine in real time. This solves the problem of "time lag bias" between prediction and observation—any prediction based on historical data will gradually deviate from the actual state over time, because the wind field in the mine may change due to fan adjustments, mine car passage, or natural air pressure fluctuations; the emissions from known release points may undergo unknown mutations; and even new unknown leak points may suddenly appear. If the window is not updated, the model will adhere to the second escape condition (random unknown source distribution) derived from outdated data, using invalid assumptions to predict the future, and the error will monotonically accumulate with the prediction time. The rolling update of the time window is equivalent to continuously "calibrating" the model with the latest measured data—whenever a new set of real wind and gas information is obtained, the time window slides forward, discarding the oldest data, incorporating the latest data, and then re-executing the optimal scenario sequence selection and overall scenario construction in step 501. This means that the set of second escape conditions (estimated distribution of unknown release sources) selected in the previous moment may be modified or even completely replaced in the new window. If new measured data shows a sudden increase in gas concentration in a certain area that cannot be explained by the original unknown sources, the optimal scenario in the new window will adjust the location or release amount of random release points to fit this change. Synchronously updating the overall scenario and the location of hazardous gas accumulation ensures that the predicted output always maintains the maximum conditional probability fit to the current real state—not simply fine-tuning old predictions with new data, but completely recalculating the optimal scenario sequence based on the updated time window, thus avoiding the error accumulation and divergence problems common in filtering methods. The rolling update mechanism gives the system the ability to track dynamic hazardous events in real time: when a real hazardous gas accumulation area is forming and migrating in a certain direction, the predicted accumulation location after each time window update will continuously move in the same direction, forming an observable trajectory; if the model detects a sudden jump or disappearance of the accumulation location, it indicates a possible drastic change in the wind field or interruption of the release source. This information can be directly used to trigger different levels of alarms or emergency responses. The output of step 502 (the updated overall scenario and aggregation location) serves as both the final prediction result for the current moment and the initial benchmark for the next time window update. This creates a closed loop that iterates over time, with each step reconstructing the model based on the latest real measurements, rather than simply extrapolating. This design elevates the problem of "hazardous gas flow direction identification and aggregation direction prediction" from a static map labeling problem to a dynamic early warning system that evolves in sync with the physical process.

[0039] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0040] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0041] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0042] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for identifying the flow direction and predicting the accumulation direction of harmful gases based on multi-sensor fusion, characterized in that, The specific steps include: Step 1: Establish a 3D model of the mine based on the morphological data of the mine, obtain the real wind force information and real gas information of each monitoring point in the mine, mark the points corresponding to each monitoring point in the 3D model of the mine to form virtual points, and associate the real wind force information and real gas information with the corresponding virtual points; Step 2: Obtain the location and air source information of the ventilation facilities in the mine, and map the location and air source information to the 3D model of the mine as input conditions. Obtain the release points and release amounts of each known harmful gas in the mine, and map the release points and release amounts to the 3D model of the mine as the first escape condition of the harmful gas. Step 3: Set multiple second escape conditions in the 3D model of the mine, and perform fluid simulation based on the input conditions, the first escape conditions and each second gas release condition to obtain multiple candidate simulation scenarios, and obtain the virtual wind force information and virtual gas information of each virtual point in each candidate simulation scenario; Step 4: Match the virtual wind and gas information in each candidate simulation scenario with the real wind and gas information respectively, and determine the candidate simulation scenario with the highest matching degree as the optimal scenario; Step 5: Continuously perform simulations and obtain real wind and gas information. Set a time window and the optimal scene sequence corresponding to each stage within the time window. Use the fluid dynamics source of the optimal scene sequence to cyclically obtain the location and concentration of harmful gases in the future.

2. The method for identifying the flow direction and predicting the accumulation direction of harmful gases based on multi-sensor fusion according to claim 1, characterized in that: Wind direction indicators and gas monitors are installed on the inner wall of the mine. Each wind direction indicator is equipped with a gas monitor. The wind direction indicator acquires the real wind force information at each location, which includes wind direction and wind force. The gas monitor acquires the real gas information at each location, which includes the content of various gases. An orthogonal coordinate system is established based on the mine, and the specific coordinates of each wind direction indicator in the orthogonal coordinate system are recorded.

3. The method for identifying the flow direction and predicting the accumulation direction of harmful gases based on multi-sensor fusion according to claim 2, characterized in that: Scan the morphological data of the inner wall of the mine and input the morphological data into the simulation software to construct a three-dimensional model of the mine. Establish the same coordinate system as the real mine in the three-dimensional model of the mine. At the same time, mark the same wind direction indicator points as in the real mine in the three-dimensional model of the mine and mark them as virtual points. Obtain virtual wind force information and virtual gas information respectively.

4. The method for identifying the flow direction and predicting the accumulation direction of harmful gases based on multi-sensor fusion according to claim 3, characterized in that: The coordinates of each wind power facility in the mine and the wind source information it generates are obtained. The wind source information includes wind force, wind direction and wind volume. The wind power facilities and their wind source information are marked in the three-dimensional model of the mine according to the coordinates and set as input conditions. The release points and release amounts of harmful gases in a real mine are obtained. The release amounts are obtained through on-site measuring instruments. At the same time, the release points and release amounts are marked in the three-dimensional model of the mine, and the points are marked as the first escape conditions.

5. The method for identifying the flow direction and predicting the accumulation direction of harmful gases based on multi-sensor fusion according to claim 4, characterized in that: Fluid simulation is performed on a 3D model of a mine shaft using simulation software. During the simulation, a second escape condition is set, and the logic of the second escape condition is as follows: At positions other than the first escape condition, release points and release amounts are set. The release amount at each release point of the second escape condition is less than the minimum value of the first escape condition. Several candidate simulation scenarios are constructed. The logic for setting release points and release amounts for the candidate simulation scenarios is as follows: A grid is defined on the surface of the 3D model of the mine. Several grid intersections are selected as escape points and escape amounts are set to form a candidate simulation scene. Several candidate simulation scenes are formed by selecting escape points and their escape amounts multiple times, and all candidate simulation scenes are made to have no completely identical escape points and escape amounts. The pressure information inside and at the entrance of a real mine is obtained, and the fluid power source is simulated by combining the input conditions, the first escape condition and the second escape condition.

6. The method for identifying the flow direction and predicting the accumulation direction of harmful gases based on multi-sensor fusion according to claim 5, characterized in that: Select the optimal scenario, including: The simulation process acquires virtual wind and gas information at each virtual point, forms a virtual information matrix based on the points, and forms a real information matrix by combining the real wind and gas information at each point in the real mine. The similarity between each virtual information matrix and the real information matrix is ​​calculated, and the simulation process corresponding to the maximum value is selected as the optimal scenario.

7. The method for identifying the flow direction and predicting the accumulation direction of harmful gases based on multi-sensor fusion according to claim 6, characterized in that: Continuously acquire real wind and gas information, and set time windows to include the latest acquired real wind and gas information. Select the corresponding optimal scenario for each time window to form an optimal scenario sequence. The input conditions, first escape condition, and second escape condition corresponding to each simulation process in the optimal scenario sequence are used as fluid power sources and cycled to form the overall scenario of this time window; The system performs future time simulations on the overall scene to obtain the predicted locations of harmful gas accumulation and their corresponding concentration values.

8. The method for identifying the flow direction and predicting the accumulation direction of harmful gases based on multi-sensor fusion according to claim 7, characterized in that: As real-time wind and gas data are updated over the time window, the overall scene and the location of harmful gas accumulation are updated simultaneously.

Images