Roadway harmful gas prediction method based on digital twinning

Abstract

The invention discloses a roadway harmful gas prediction method based on digital twinning, and relates to the technical field of intelligent mines, and the method comprises the steps: building a physical roadway, and carrying out the data collection; constructing a roadway harmful gas prediction model under the condition that the number of sensors is limited based on a physical information neural network; the method specifically comprises the steps of data set division, physical constraint selection, PINN parameter setting, automatic differential calculation, PINN parameter optimization and general neural network agent model export. Performing prediction through agent model calculation; through a tetrahedral mesh, construction of a virtual entity and rendering updating, the method is based on a physical information neural network and construction of a digital twin proxy model at the same time, the concentration of harmful gas in a roadway space can be comprehensively predicted under the condition that the number of sensors is limited, and virtual entity rendering updating is achieved through data transmission. The problem that a roadway blank area cannot be predicted due to the fact that the number of sensors is limited can be solved.

Description

Technical Field

[0001] This invention relates to the field of smart mining technology, specifically a method for predicting hazardous gases in roadways based on digital twins. Background Technology

[0002] Coal, as one of my country's important basic energy sources, has long occupied a dominant position in the energy production and consumption structure, and its abundant reserves play a vital supporting role in national energy security and economic development. However, because my country's coal resources are mostly located in deep strata, coal mining is generally carried out underground, resulting in a complex, enclosed working environment and high safety risks. In underground coal mine production, the mine ventilation system plays a crucial safety role. Its main functions include continuously supplying fresh air to various underground work sites, providing sufficient oxygen supply for workers, diluting and removing toxic and harmful gases such as methane and coal dust, improving the underground working environment, and regulating underground temperature and humidity conditions. It is an essential infrastructure for ensuring safe production in underground coal mines.

[0003] Currently, in most coal mine production practices, underground ventilation systems not only suffer from uneven distribution of monitoring points and limited coverage, easily creating monitoring blind spots, but also exhibit significant deficiencies in data acquisition efficiency, real-time performance, and system intelligence. Furthermore, when temporary anomalies in harmful gases or sudden changes in operating conditions occur, traditional methods often struggle to identify problems promptly and accurately and take effective measures, posing a potential threat to mine safety.

[0004] Against this backdrop, digital twin technology offers a novel approach to intelligent coal mine ventilation. By constructing a digital model in virtual space that closely mirrors the physical tunnel environment, and combining it with monitoring data and intelligent algorithms, the digital twin system can predict dynamic trends in the tunnel environment, providing support for optimizing ventilation system operations and making safety decisions. However, in practical applications, limitations such as sensor deployment conditions, costs, and the complex underground environment mean that comprehensively and accurately acquiring the status information of each node in the tunnel remains one of the key scientific and engineering problems hindering the in-depth application of digital twin technology in the coal mining sector. Summary of the Invention

[0005] This invention provides a method for predicting hazardous gases in roadways based on digital twins, which solves the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A method for predicting hazardous gases in roadways based on digital twins includes the following steps:

[0008] Step 1: Construct a physical tunnel and collect data, mainly including tunnel support structure, wall structure, tunnel geometry, spatial features, and sensor grid deployment;

[0009] Step 2: Based on physical information neural network prediction, under the condition of a limited number of sensors, a dataset is constructed by collecting tunnel environment data, the gas diffusion law is analyzed, differential equation physical constraints are selected, parameters are set in combination with deep learning laws, automatic differential input and output are completed, physical constraint loss is obtained, network parameters are optimized according to gradient, and the set training conditions are verified.

[0010] Step 3: Export the network model obtained in Step 2 as a general open standard network model, and deploy the model as a proxy model for the digital twin to perform prediction tasks;

[0011] Step 4: Virtual entity construction. A digital twin model is constructed using the tetrahedral mesh method.

[0012] Step 5: Virtual entity rendering update. Select the color model and update the rendering using normalized prediction data.

[0013] As a preferred embodiment of the present invention, step 1 specifically includes:

[0014] S11: Construct physical tunnel supports based on the actual tunnel geometry, spatial structure, and cross-sectional features;

[0015] S12: Seal off the supports to complete the construction of the physical conditions of the wall;

[0016] S13: Connects different tunnels and constructs a corner structure for physical entities.

[0017] As a preferred embodiment of the present invention, step 2 specifically includes:

[0018] S21: Deploy a limited number of sensors in the physical tunnel to collect environmental data of the physical tunnel and build a dataset;

[0019] S22: Based on fluid mechanics, the diffusion law of harmful gases is analyzed, and a suitable PINN physical constraint is selected. The physical constraint includes the influence of wind force, and the diffusion of harmful gases requires the introduction of a convection mechanism. Therefore, the convection-diffusion equation for the propagation of harmful gases in the airflow field is:

[0020]

[0021] Where c is the gas concentration, x, y, z are the three-dimensional spatial coordinates, and D is the diffusion coefficient;

[0022] Due to the presence of harmful gas release sources, and the need to modify equation (1), a diffusion source mechanism needs to be introduced, resulting in the three-dimensional convection-diffusion-source term equation:

[0023]

[0024] Where Sc is the gas source term;

[0025] S23: Select network parameters based on commonly used neural network parameter settings, combined with engineering practice and experience;

[0026] S24: Automatic differentiation is a technique for accurately and efficiently calculating derivatives in a computer. The partial derivatives of equation (2) can be realized through automatic differentiation to complete the construction of the physical constraint loss function. S25: Optimize network parameters by gradient descent through physical constraint loss and data loss, guided by neural network training;

[0027] S26: Determine whether the requirements are met by judging the conditions, and end the training early to save computing resources;

[0028] S27: Select an open standard neural network, derive the surrogate model, and complete the training of the physical information neural network.

[0029] As a preferred embodiment of the present invention, step 3 specifically includes:

[0030] S31: Proxy model deployment, deploying the ONNX format model on the digital twin platform;

[0031] S32: Using spatiotemporal input and a limited number of sensor data, perform network model calculations to obtain prediction results.

[0032] As a preferred embodiment of the present invention, step 4 specifically includes:

[0033] S41: Create a 3D model of the solid tunnel at the same scale and export it in obj format;

[0034] S42: Use HyperMesh to generate tetrahedral meshes, construct virtual entities, and export inp format;

[0035] S43: Read the inp file through the digital twin platform and construct a digital twin model.

[0036] As a preferred embodiment of the present invention, step 5 specifically includes:

[0037] S51: Normalize the predicted data to obtain the 0-1 data standard;

[0038] S52: Select the HSV color space model and calculate the corresponding hue;

[0039] S53: Perform 3D rendering based on hue, saturation, and brightness.

[0040] The present invention has the following advantages:

[0041] 1. A surrogate model of the roadway environment is constructed by introducing a Physical Information Neural Network (PINN). This model embeds physical constraints such as gas diffusion into the neural network training process, enabling it to accurately learn the spatiotemporal evolution of roadway environmental parameters even with limited monitoring data. By constructing a surrogate model based on PINN, the prediction gap problem caused by the limited distribution of sensors in underground coal mines can be significantly alleviated, achieving continuous and comprehensive perception of roadway environmental information. This provides more comprehensive and reliable data support for mine ventilation safety management, gas disaster early warning, and real-time updates of digital twin systems.

[0042] 2. By constructing virtual entity models that are highly consistent with physical entities, the state changes of physical entities can be intuitively displayed in virtual space. Virtual entities, through receiving predictive data, achieve dynamic mapping of the spatiotemporal evolution process of physical entities, enabling changes in the physical system to be presented in a visual form within the digital environment.

[0043] 3. By introducing the HSV color space model, the predicted tunnel environmental parameters are mapped into the digital twin virtual tunnel model in the form of a visual cloud map. By normalizing the environmental information and establishing a correspondence between it and the hue, saturation and brightness in the HSV color space, the numerical changes of environmental parameters can be transformed into intuitive color gradient changes, thereby clearly showing the spatial distribution characteristics and changing trends of environmental information in the tunnel. Attached Figure Description

[0044] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0045] Figure 1 This is a schematic diagram of the structure of the present invention;

[0046] Figure 2 This is a diagram of the physical information neural method of the present invention;

[0047] Figure 3 This is the digital twin virtual entity diagram of the present invention;

[0048] Figure 4 This is a comparison diagram of the physical information neural network prediction of the present invention;

[0049] Figure 5 This is a digital twin rendering of the present invention;

[0050] Figure 6 This is a cross-sectional view of the digital twin of the present invention;

[0051] Figure 7 This is a flowchart of the present invention. Detailed Implementation

[0052] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0053] It should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the present invention.

[0054] For examples, please refer to Figures 1-7 A method for predicting hazardous gases in roadways based on digital twins includes the following steps:

[0055] Step 1: Construct a physical tunnel and collect data, mainly including the tunnel support structure, wall structure, tunnel geometry, spatial features, and sensor grid deployment, etc. Specific details include the following:

[0056] S11: Construct physical tunnel supports based on the actual tunnel geometry, spatial structure, and cross-sectional features;

[0057] S12: Seal off the supports to complete the construction of the physical conditions of the wall;

[0058] S13: Connects different tunnels and constructs a corner structure for physical entities.

[0059] Step 2: Based on physical information neural network prediction, under the condition of a limited number of sensors, a dataset is constructed by collecting tunnel environmental data. The gas diffusion law is analyzed, physical constraints of differential equations are selected, and parameters are set in combination with deep learning principles. Automatic differential input and output are completed, and physical constraint loss is obtained. The network parameters are optimized according to the gradient, and the set training conditions are verified. The specific content includes the following:

[0060] S21: Deploy a limited number of sensors in the physical tunnel to collect environmental data of the physical tunnel and build a dataset;

[0061] S22: Based on fluid mechanics, the diffusion law of harmful gases is analyzed, and a suitable PINN physical constraint is selected. The physical constraint includes the influence of wind force, and the diffusion of harmful gases requires the introduction of a convection mechanism. Therefore, the convection-diffusion equation for the propagation of harmful gases in the airflow field is:

[0062]

[0063] Where c is the gas concentration, x, y, z are the three-dimensional spatial coordinates, and D is the diffusion coefficient;

[0064] Due to the presence of harmful gas release sources, and the need to modify equation (1), a diffusion source mechanism needs to be introduced, resulting in the three-dimensional convection-diffusion-source term equation:

[0065]

[0066] Where Sc is the gas source term;

[0067] S23: Select network parameters based on commonly used neural network parameter settings, combined with engineering practice and experience;

[0068] S24: Automatic differentiation is a technique for accurately and efficiently calculating derivatives in a computer. The partial derivatives of equation (2) can be realized through automatic differentiation to complete the construction of the physical constraint loss function. S25: Optimize network parameters by gradient descent through physical constraint loss and data loss, guided by neural network training;

[0069] S26: Determine whether the requirements are met by judging the conditions, and end the training early to save computing resources;

[0070] S27: Select an open standard neural network, derive the surrogate model, and complete the training of the physical information neural network.

[0071] Step 3: Export the network model obtained in Step 2 as a general open standard network model. Deploy the model as a proxy model for the digital twin to perform prediction tasks. The specific content includes the following:

[0072] S31: Proxy model deployment, deploying the ONNX format model on the digital twin platform;

[0073] S32: Using spatiotemporal input and a limited number of sensor data, perform network model calculations to obtain prediction results.

[0074] Step 4: Virtual entity construction. A digital twin model is constructed using the tetrahedral mesh method. Specific details include the following:

[0075] S41: Create a 3D model of the solid tunnel at the same scale and export it in obj format;

[0076] S42: Use HyperMesh to generate tetrahedral meshes, construct virtual entities, and export inp format;

[0077] S43: Read the inp file through the digital twin platform and construct a digital twin model.

[0078] Step 5: Virtual entity rendering update. Select the color model and perform rendering updates using normalized prediction data. Specific details include:

[0079] S51: Normalize the predicted data to obtain the 0-1 data standard;

[0080] S52: Select the HSV color space model and calculate the corresponding hue;

[0081] S53: Perform 3D rendering based on hue, saturation, and brightness.

[0082] Figures 4-6 The final results of this embodiment are shown. This embodiment deploys a total of 16 gas sensors, with data from 15 sensors used as training and validation datasets, and data from the sensor closest to the inlet used for validation comparison.

[0083] like Figure 4 As shown, the actual data and predicted data are plotted and compared. The blue curve represents the model's predicted output at the spatial node, while the red curve represents the actual value measured by the sensor. The two curves show a high degree of consistency in trend over the entire time range, with an absolute percentage error of 2.87%. This indicates that even with limited observation data and complex environmental conditions, PINN can still predict the spatiotemporal evolution of gas concentration with high accuracy, demonstrating strong practical application value and engineering applicability.

[0084] like Figure 5 As shown, the virtual tunnel model is rendered in three dimensions using the HSV color space model and combined with three-dimensional rendering technology to perform coloring and visualization processing, thereby obtaining a digital twin at the initial moment when the gas content is zero and appears blue.

[0085] like Figure 6As shown, by setting a cutting plane at a specific location, the three-dimensional tunnel model is cut along a specified direction, thereby obtaining a two-dimensional cross-sectional display of the tunnel's internal environmental information. This cross-sectional view can clearly reflect the numerical changes of environmental parameters at different locations within the tunnel, highlighting key features such as concentration gradients, local accumulation areas, and diffusion paths.

[0086] This method achieves comprehensive prediction of hazardous gases in roadways under the condition of a limited number of sensors through a physical information neural network. A digital twin virtual model is constructed, and the HSV color space model is introduced. Through predictive data transmission, the virtual roadway is updated, intuitively displaying hazardous gas information.

[0087] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting hazardous gases in roadways based on digital twins, characterized in that, Includes the following steps: Step 1: Construct a physical tunnel and collect data, mainly including tunnel support structure, wall structure, tunnel geometry, spatial features, and sensor grid deployment; Step 2: Based on physical information neural network prediction, under the condition of a limited number of sensors, a dataset is constructed by collecting tunnel environment data, the gas diffusion law is analyzed, differential equation physical constraints are selected, parameters are set in combination with deep learning laws, automatic differential input and output are completed, physical constraint loss is obtained, network parameters are optimized according to gradient, and the set training conditions are verified. Step 3: Export the network model obtained in Step 2 as a general open standard network model, and deploy the model as a proxy model for the digital twin to perform prediction tasks; Step 4: Virtual entity construction. A digital twin model is constructed using the tetrahedral mesh method. Step 5: Virtual entity rendering update. Select the color model and update the rendering using normalized prediction data.

2. The method for predicting hazardous gases in roadways based on digital twins according to claim 1, characterized in that, Step 1 specifically includes: S11: Construct physical tunnel supports based on the actual tunnel geometry, spatial structure, and cross-sectional features; S12: Seal off the supports to complete the construction of the physical conditions of the wall; S13: Connects different tunnels and constructs a corner structure for physical entities.

3. The method for predicting hazardous gases in roadways based on digital twins according to claim 1, characterized in that, Step 2 specifically includes: S21: Deploy a limited number of sensors in the physical tunnel to collect environmental data of the physical tunnel and build a dataset; S22: Based on fluid mechanics, the diffusion law of harmful gases is analyzed, and a suitable PINN physical constraint is selected. The physical constraint includes the influence of wind force, and the diffusion of harmful gases requires the introduction of a convection mechanism. Therefore, the convection-diffusion equation for the propagation of harmful gases in the airflow field is: ， Where c is the gas concentration, x, y, z are the three-dimensional spatial coordinates, and D is the diffusion coefficient; Due to the presence of harmful gas release sources, and the need to modify equation (1), a diffusion source mechanism needs to be introduced, resulting in the three-dimensional convection-diffusion-source term equation: ， Where Sc is the gas source term; S23: Select network parameters based on commonly used neural network parameter settings, combined with engineering practice and experience; S24: Automatic differentiation is a technique for accurately and efficiently calculating derivatives in a computer. The partial derivatives of equation (2) can be realized through automatic differentiation to complete the construction of the physical constraint loss function. S25: Optimize network parameters by gradient descent through physical constraint loss and data loss, guided by neural network training; S26: Determine whether the requirements are met by judging the conditions, and end the training early to save computing resources; S27: Select an open standard neural network, derive the surrogate model, and complete the training of the physical information neural network.

4. The method for predicting hazardous gases in roadways based on digital twins according to claim 1, characterized in that, Step 3 specifically includes: S31: Proxy model deployment, deploying the ONNX format model on the digital twin platform; S32: Using spatiotemporal input and a limited number of sensor data, perform network model calculations to obtain prediction results.

5. The method for predicting hazardous gases in roadways based on digital twins according to claim 1, characterized in that, Step 4 specifically includes: S41: Create a 3D model of the solid tunnel at the same scale and export it in obj format; S42: Use HyperMesh to generate tetrahedral meshes, construct virtual entities, and export inp format; S43: Read the inp file through the digital twin platform and construct a digital twin model.

6. The method for predicting hazardous gases in roadways based on digital twins according to claim 1, characterized in that, Step 5 specifically includes: S51: Normalize the predicted data to obtain the 0-1 data standard; S52: Select the HSV color space model and calculate the corresponding hue; S53: Perform 3D rendering based on hue, saturation, and brightness.

Images