PINN-based coal mine wind network sparse wind resistance inversion modeling method and system

Abstract

The invention discloses a PINN-based coal mine wind network sparse wind resistance inversion modeling method and system, and relates to the technical field of coal mine wind network sparse wind resistance inversion modeling, and the method comprises the steps: obtaining a node and roadway topological structure of a coal mine ventilation network, and combining roadway attribute parameters and air volume, wind pressure and temperature multi-mode sensor data to obtain a wind resistance inversion model; constructing graph structure input data comprising a node feature matrix, an edge feature matrix and an adjacent matrix; a GNN-PINN hybrid model is established based on the graph structure input data, a graph convolutional layer is utilized to extract wind network topology correlation features, and the effect of residual branch chords in wind volume optimization is enhanced through a side attention mechanism; in the hybrid model, according to a minimum spanning tree decomposition result, the air volume of a roadway in a spanning tree is directly calculated from residual branch chord air volume variables through correlation coefficient linear combination, and hard constraint output which does not participate in gradient updating is formed; and calculating the air pressure balance residual error of each independent loop based on the air volume output by the hard constraint and the air resistance predicted by the model.

Description

Technical Field

[0001] This invention relates to the field of coal mine ventilation network sparseness resistance inversion modeling technology, and in particular to a coal mine ventilation network sparseness resistance inversion modeling method and system based on PINN. Background Technology

[0002] Coal mine ventilation network sparseness resistance inversion modeling technology is a hybrid modeling method combining graph neural networks (GNNs) and physical information neural networks (PINNs) to invert and estimate the resistance parameters of each roadway in the coal mine ventilation network. Therefore, how to utilize advanced technologies to improve the intelligence level and safety of coal mine ventilation network sparseness resistance inversion modeling has become one of the urgent problems to be solved.

[0003] In the field of sparse wind resistance inversion modeling of coal mine ventilation networks, traditional methods often rely on a large amount of dense air volume measurement data. However, in actual coal mine environments, due to the high cost of sensor deployment and the difficulty of maintenance, air volume data is sparse and incomplete. Furthermore, traditional methods are difficult to effectively identify and locate abnormal roadways in the ventilation network. At the same time, traditional inversion methods often exhibit poor generalization ability when faced with complex and dynamically changing ventilation networks. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a PINN-based method for sparse wind resistance inversion modeling of coal mine ventilation networks. This addresses the problem that traditional methods often rely on a large amount of dense air volume measurement data. However, in actual coal mine environments, due to the high cost of sensor deployment and maintenance difficulties, air volume data is sparse and incomplete. Furthermore, traditional methods struggle to effectively identify and locate abnormal roadways in the ventilation network. At the same time, traditional inversion methods often exhibit poor generalization ability when faced with complex and dynamically changing ventilation networks.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for inverting and modeling wind resistance in coal mine ventilation networks based on PINN, which includes: Obtain the node and roadway topology of the coal mine ventilation network, and construct a graph structure input data including node feature matrix, edge feature matrix and adjacency matrix by combining roadway attribute parameters and multimodal sensor data of air volume, air pressure and temperature. A GNN-PINN hybrid model is established based on the graph structure input data. The topological association features of the wind network are extracted using graph convolutional layers, and the role of co-branch strings in wind volume optimization is enhanced through the edge attention mechanism. In the hybrid model, based on the minimum spanning tree decomposition results, the air volume of the roadway within the spanning tree is directly calculated by the co-branch chord air volume variables through a linear combination of correlation coefficients, forming a hard constraint output that does not participate in gradient updates. Based on the air volume output by the hard constraints and the wind resistance predicted by the model, the wind pressure balance residual and the roadway pressure drop physical relationship residual of each independent loop are calculated, and the fitting residual is constructed by combining the multimodal sensor data to form a soft physical loss term. An adaptive sparsification module is constructed to generate personalized L1 regularization coefficients and dynamic truncation thresholds for each roadway based on roadway type, historical volatility and spatial location. These are used to construct the sparsity loss term and control the truncation operation of wind resistance change. The soft physical loss term and the sparse loss term are combined to form the total loss function, which drives the training of the hybrid model, outputs the inverted wind resistance value of each roadway, and locates abnormal roadways based on the comparison results of wind resistance change and dynamic threshold.

[0007] As a preferred embodiment of the PINN-based coal mine wind network sparse wind resistance inversion modeling method described in this invention, the minimum spanning tree decomposition result specifically includes the following steps: The physical structure of the coal mine ventilation network is abstracted as an undirected connected graph, which consists of a set of nodes and a set of edges. Each element in the set of nodes represents a key connection point in the ventilation network, and each element in the set of edges represents a roadway connecting two nodes. Assign a weight to each edge in the edge set. The value of the weight is the absolute value of the historical air volume data recorded for the roadway under normal and stable ventilation conditions. The minimum spanning tree algorithm is used to process undirected connected graphs to find a subgraph that contains all nodes in the set of nodes. This subgraph satisfies two conditions: the subgraph itself does not contain any cycles and the sum of the weights of all edges in the subgraph is the minimum value among all possible spanning trees. Define the subgraph obtained by the above solution as a spanning tree; The set of all edges in the original edge set that do not belong to the spanning tree is defined as the co-branch chord set.

[0008] As a preferred embodiment of the PINN-based coal mine ventilation network sparseness wind resistance inversion modeling method described in this invention, the specific steps for directly calculating the air volume of the roadway within the generator tree from the co-branch chord air volume variables through a linear combination of correlation coefficients are as follows: For each co-chord in the set of co-chords, determine the unique basic cycle formed by the co-chord and the edges in the spanning tree; The basic loop is constructed by starting from one endpoint of the co-branch, following the path in the spanning tree to its other endpoint, and then returning to the starting point through the co-branch itself, forming a closed loop. In the basic circuit, all edges except co-chords are identified. All edges belong to the spanning tree and are denoted as spanning tree edges. Define a unified basic loop traversal direction, wherein the basic loop traversal direction is either clockwise or counterclockwise; For each spanning tree edge in the circuit, compare its preset edge direction in the undirected connected graph with the circuit traversal direction to see if they are consistent. If the direction of the spanning tree edge is the same as the direction of the circuit traversal, then set its correlation coefficient with the co-chord to +A; If the direction of the generated tree edge is opposite to the direction of the circuit traversal, then its correlation coefficient with the co-chord is set to -A.

[0009] As a preferred embodiment of the PINN-based coal mine wind network sparseness wind resistance inversion modeling method described in this invention, the wind pressure balance residual specifically comprises the following steps: For each basic loop generated by a co-branch chord, list all the lane edges contained in the loop; For each edge in the loop, calculate its pressure drop based on the edge's wind resistance and the wind volume calculated from the hard constraints output by the model. The expression is: ; in, The inversion drag value of the edge, The airflow value at the edge; Determine the direction coefficient of the edge based on its direction in the loop; If the direction of the edge is the same as the traversal direction of the circuit, the direction coefficient is +B; if the direction is opposite, the direction coefficient is -B. Calculate the algebraic sum of wind pressure for this basic loop. The expression is: ; The summation operation covers all edges in the circuit. Direction coefficient; Will The square of the value is used as the wind pressure balance residual term for this loop.

[0010] As a preferred embodiment of the PINN-based coal mine ventilation network sparseness wind resistance inversion modeling method described in this invention, the specific steps for the residual physical relationship of roadway pressure drop are as follows: For each edge in the ventilation network, read the real-time temperature value measured by its associated temperature sensor; Calculate the air density of the region containing the edge using the ideal gas law and the measured temperature. The expression is: ; in, This refers to the local atmospheric pressure. is the gas constant of air; Obtain the initial wind resistance value of this side, and correct it to the equivalent wind resistance under the current environment based on the calculated air density; ; in, This refers to the air density under standard conditions. Using the equivalent wind resistance and the air volume output by hard constraints, the theoretical wind pressure drop on this side is calculated using the following formula: ; If a wind pressure sensor is deployed on this side, the wind pressure drop value measured by it will be obtained; Calculate the pressure drop deviation on this side: when a measured value exists, the deviation is the difference between the theoretical wind pressure drop and the measured wind pressure drop; When no measured value is available, the deviation is the difference between the theoretical wind pressure drop and the wind pressure drop calculated by the model; The square of the deviation is used as the residual term of the physical relationship of the pressure drop on that side.

[0011] As a preferred embodiment of the PINN-based coal mine wind network sparse wind resistance inversion modeling method described in this invention, the personalized L1 regularization coefficient and the dynamic truncation threshold are specifically implemented as follows: For each lane side, input its lane type identifier, the standard deviation of historical air volume data, and the straight-line distance from the side to the main ventilation fan; Determine the basic sensitivity based on the roadway type identification; The formula for calculating the impact factor of historical air volume fluctuations is as follows: ; in, This represents the maximum standard deviation of historical airflow across all edges in the entire ventilation network. The formula for calculating the spatial distance influence factor is as follows: ; in, This is a preset distance constant; The overall anomaly sensitivity is calculated using the following formula: ; Set the personalized L1 regularization coefficient for this edge. Dynamic truncation threshold ,in, and These are the preset baseline parameters.

[0012] As a preferred embodiment of the PINN-based coal mine ventilation network sparseness wind resistance inversion modeling method described in this invention, the specific steps of the wind resistance change truncation operation are as follows: Set the execution cycle of the truncation operation to every The model training is executed once per iteration. It is a positive integer; If the current training iteration count is When the value is an integer multiple of the specified value, the truncation process is initiated. For each lane, read the change in wind resistance output by the model in the current training iteration and its corresponding dynamic cutoff threshold. Determine whether the absolute value of the change in wind resistance is less than the dynamic cutoff threshold; If the judgment result is yes, then the change in wind resistance on that side will be forcibly set to 0; Mark the edge where the wind resistance change is set to 0 as frozen; In subsequent training iterations, for edges that are frozen, the change in wind resistance remains 0, and they do not participate in gradient calculation and parameter updates until the model training process terminates.

[0013] Secondly, this invention provides a PINN-based coal mine ventilation network sparsity inversion modeling system for wind resistance, comprising: The module includes a graph data construction module, a model building module, a residual calculation module, a sparsity control module, a loss optimization module, and a result output module. The graph data construction module is used to obtain the node and roadway topology of the coal mine ventilation network, and combine roadway attribute parameters and multi-modal sensor data of air volume, air pressure and temperature to construct graph structure input data including node feature matrix, edge feature matrix and adjacency matrix. The model building module is used to build a GNN-PINN hybrid model based on the graph structure input data, extract the topological correlation features of the wind network using graph convolutional layers, and strengthen the role of co-branch string in wind volume optimization through edge attention mechanism; according to the minimum spanning tree decomposition results, the wind volume of the roadway in the spanning tree is directly calculated by the co-branch string wind volume variables through linear combination of correlation coefficients, forming a hard constraint output that does not participate in gradient update. The residual calculation module is used to calculate the wind pressure balance residual and the roadway pressure drop physical relationship residual of each independent loop based on the wind volume output by hard constraints and the wind resistance predicted by the model, and to construct the fitting residual by combining the multimodal sensor data to form a soft physical loss term. The sparse control module is used to generate personalized L1 regularization coefficients and dynamic truncation thresholds for each roadway based on roadway type, historical volatility and spatial location, to form sparse loss terms and control the truncation operation of wind resistance change. The loss optimization module is used to combine the soft physical loss term and the sparse loss term to form a total loss function, driving the training of the hybrid model; The result output module is used to output the inverted wind resistance value of each roadway and locate abnormal roadways based on the comparison result of wind resistance change and dynamic threshold.

[0014] Thirdly, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the PINN-based coal mine wind network sparseness wind resistance inversion modeling method as described in the first aspect of the present invention.

[0015] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the PINN-based coal mine wind network sparse resistance inversion modeling method as described in the first aspect of the present invention.

[0016] The beneficial effects of this invention are as follows: By integrating graph neural networks and physical information neural networks, a sparse wind resistance inversion model for coal mine ventilation networks is constructed. The minimum spanning tree decomposition and correlation coefficient mechanism are used to realize the hard constraint calculation of air volume, effectively ensuring the physical laws of air volume conservation and wind pressure balance. By introducing multimodal sensor data and adaptive sparsity control strategy, the accuracy of wind resistance inversion and the generalization ability of the model are significantly improved. It can accurately identify abnormal roadways and provide reliable technical support for the safe and efficient operation of coal mine ventilation systems. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of the coal mine wind resistance inversion modeling method based on PINN in Example 1.

[0019] Figure 2 This is a schematic diagram of the coal mine wind resistance inversion modeling system based on PINN in Example 1. Detailed Implementation

[0020] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0021] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0022] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0023] Example, refer to Figure 1 and Figure 2 This embodiment of the invention provides a method for inverting and modeling wind resistance in a coal mine ventilation network based on PINN, comprising the following steps: S1. Obtain the node and roadway topology of the coal mine ventilation network, and combine roadway attribute parameters and multi-modal sensor data of air volume, air pressure, and temperature to construct graph structure input data including node feature matrix, edge feature matrix and adjacency matrix; Furthermore, the feature vector of each node in the node feature matrix includes the altitude of the connection point, the local ventilation resistance coefficient, and the relative position coordinates with the main ventilation fan; the feature vector of each lane in the edge feature matrix includes its cross-sectional area, perimeter, friction resistance coefficient, length, and real-time air volume, air pressure, and temperature data collected by sensors deployed on the lane; the adjacency matrix uses a sparse storage format to represent the connection relationship of the ventilation network, where a matrix element of 1 indicates that there is a directly connected lane between two nodes, and 0 indicates that there is no direct connection. It should be noted that when constructing graph structure input data, in addition to ensuring the spatiotemporal consistency and reliability of the data, the possibility of data loss or sensor failure in the roadway network should also be considered. Using appropriate data interpolation methods and anomaly detection algorithms is crucial to improving the quality of model input data. Considering the complexity of the coal mine ventilation environment, the parameters in the node and edge feature matrices need to be updated regularly to reflect the latest ventilation network status.

[0024] S2. Based on graph structure input data, a GNN-PINN hybrid model is established. The topological association features of the wind network are extracted using graph convolutional layers. The role of co-branch strings in wind volume optimization is enhanced through the edge attention mechanism. Furthermore, the graph convolutional layer adopts a multi-layer stacked structure. Each layer updates the embedding representation of the current node by aggregating the feature information of neighboring nodes, thereby achieving deep feature extraction of the global topology of the wind network. The edge attention mechanism is a learnable weight allocation module. Its input is the feature vector of the co-branch and its associated generated tree edge. By calculating the attention score, the contribution of each edge in the wind volume distribution is dynamically adjusted, so that the model pays more attention to the key co-branch that has a significant impact on the wind volume distribution. It should be noted that the design of the GNN-PINN hybrid model not only relies on accurate graph structure input data, but also needs to be combined with specific physical rules to ensure that the physical meaning of the model prediction results is reasonable. At the same time, by introducing the edge attention mechanism, the model can adaptively adjust the influence weight of different co-branch chords on airflow optimization, thereby achieving effective identification and optimization of critical paths in complex ventilation networks.

[0025] S3. In the hybrid model, based on the minimum spanning tree decomposition results, the air volume of the roadway within the spanning tree is directly calculated by the co-branch chord air volume variables through a linear combination of correlation coefficients, forming a hard constraint output that does not participate in gradient updates. Furthermore, the minimum spanning tree decomposition result, specifically the following steps: The physical structure of the coal mine ventilation network is abstracted as an undirected connected graph, which consists of a set of nodes and a set of edges. Each element in the set of nodes represents a key connection point in the ventilation network, and each element in the set of edges represents a roadway connecting two nodes. Assign a weight to each edge in the edge set. The value of the weight is the absolute value of the historical air volume data recorded for the roadway under normal and stable ventilation conditions. The minimum spanning tree algorithm is used to process undirected connected graphs to find a subgraph that contains all nodes in the set of nodes. This subgraph satisfies two conditions: the subgraph itself does not contain any cycles and the sum of the weights of all edges in the subgraph is the minimum value among all possible spanning trees. Define the subgraph obtained by the above solution as a spanning tree; The set of all edges in the original edge set that do not belong to the spanning tree is defined as the cobranch set; For each co-chord in the set of co-chords, determine the unique basic cycle formed by the co-chord and the edges in the spanning tree; The basic loop is constructed by starting from one endpoint of the co-curve, following the path in the spanning tree to its other endpoint, and then returning to the starting point through the co-curve itself, forming a closed loop. In the basic circuit, all edges except co-chords are identified. All edges belong to the spanning tree and are denoted as spanning tree edges. Define a uniform basic circuit traversal direction, which can be either clockwise or counterclockwise. For each spanning tree edge in the circuit, compare its preset edge direction in the undirected connected graph with the circuit traversal direction to see if they are consistent. If the direction of the spanning tree edge is the same as the direction of the circuit traversal, then set its correlation coefficient with the co-chord to +A; If the direction of the spanning tree edge is opposite to the direction of the circuit traversal, then its correlation coefficient with the co-chord is set to -A; It should be noted that the results of minimum spanning tree decomposition not only provide a basic framework for subsequent air volume calculation, but also make the hard constraint output more consistent with the actual operation of ventilation networks by accurately modeling the relationship between the edges and co-branch chords of the spanning tree. Setting a unified basic loop traversal direction helps to simplify the calculation process, ensure the consistency of correlation coefficients, and thus improve the accuracy of air volume calculation.

[0026] S4. Based on the air volume output by hard constraints and the wind resistance predicted by the model, calculate the wind pressure balance residual and the physical relationship residual of roadway pressure drop for each independent loop, and combine them with multimodal sensor data to construct fitting residuals, which together form a soft physical loss term. Furthermore, the wind pressure balance residual is determined through the following steps: For each basic loop generated by a co-branch chord, list all the lane edges contained in the loop; For each edge in the loop, calculate its pressure drop based on the edge's wind resistance and the wind volume calculated from the hard constraints output by the model. The expression is: ; in, The inversion drag value of the edge, The airflow value at the edge; Determine the direction coefficient of the edge based on its direction in the loop; If the direction of the edge is the same as the traversal direction of the circuit, the direction coefficient is +B; if the direction is opposite, the direction coefficient is -B. Calculate the algebraic sum of wind pressure for this basic loop. The expression is: ; The summation operation covers all edges in the circuit. Direction coefficient; Will The square of the value is used as the wind pressure balance residual term for this loop; The physical relationship residual of roadway pressure drop, the specific steps are as follows: For each edge in the ventilation network, read the real-time temperature value measured by its associated temperature sensor; Calculate the air density of the region containing the edge using the ideal gas law and the measured temperature. The expression is: ; in, This refers to the local atmospheric pressure. is the gas constant of air; Obtain the initial wind resistance value of this side, and correct it to the equivalent wind resistance under the current environment based on the calculated air density; ; in, This refers to the air density under standard conditions. Using the equivalent wind resistance and the air volume output by hard constraints, the theoretical wind pressure drop on this side is calculated using the following formula: ; If a wind pressure sensor is deployed on this side, the wind pressure drop value measured by it will be obtained; Calculate the pressure drop deviation on this side: when a measured value exists, the deviation is the difference between the theoretical wind pressure drop and the measured wind pressure drop; When no measured value is available, the deviation is the difference between the theoretical wind pressure drop and the wind pressure drop calculated by the model; The square of the deviation is used as the residual term of the physical relationship of the pressure drop on that side; It should be noted that the construction of the soft physical loss term takes into account the differences between theoretical calculations and actual measurements. By minimizing these residuals to drive model training, the prediction accuracy of the model can be effectively improved. When dealing with the residuals of the physical relationship of roadway pressure drop, real-time temperature data is used to correct the wind resistance, which further enhances the model's adaptability to environmental changes and enables it to more accurately reflect the actual operating status of the ventilation network.

[0027] S5. Construct an adaptive sparsification module to generate personalized L1 regularization coefficients and dynamic truncation thresholds for each roadway based on roadway type, historical volatility and spatial location. These are used to construct sparsity loss terms and control the truncation operation of wind resistance changes. Furthermore, the personalized L1 regularization coefficient and dynamic truncation threshold are defined by the following steps: For each lane side, input its lane type identifier, the standard deviation of historical air volume data, and the straight-line distance from the side to the main ventilation fan; Determine the basic sensitivity based on the roadway type identification; The formula for calculating the impact factor of historical air volume fluctuations is as follows: ; in, This represents the maximum standard deviation of historical airflow across all edges in the entire ventilation network. The formula for calculating the spatial distance influence factor is as follows: ; in, This is a preset distance constant; The overall anomaly sensitivity is calculated using the following formula: ; Set the personalized L1 regularization coefficient for this edge. Dynamic truncation threshold ,in, and These are pre-set baseline parameters; It should be noted that the introduction of personalized L1 regularization coefficients and dynamic truncation thresholds aims to balance model complexity and generalization ability, avoid overfitting while retaining important feature information. Adjusting the regularization strength according to factors such as roadway type, historical volatility and spatial location can make the model more targeted and flexible, especially suitable for coal mine ventilation networks with high heterogeneity.

[0028] S6. Combine the soft physical loss term and the sparse loss term to form the total loss function, drive the training of the hybrid model, output the inverted wind resistance value of each roadway, and locate abnormal roadways based on the comparison results of wind resistance change and dynamic threshold. Furthermore, the specific steps for truncating changes in wind resistance are as follows: Set the execution cycle of the truncation operation to every The model training is executed once per iteration. It is a positive integer; If the current training iteration count is When the value is an integer multiple of the specified value, the truncation process is initiated. For each lane, read the change in wind resistance output by the model in the current training iteration and its corresponding dynamic cutoff threshold. Determine whether the absolute value of the change in wind resistance is less than the dynamic cutoff threshold; If the judgment result is yes, then the change in wind resistance on that side will be forcibly set to 0; Mark the edge where the wind resistance change is set to 0 as frozen; In subsequent training iterations, for edges that are in a frozen state, the change in wind resistance remains at 0, and they do not participate in gradient calculation and parameter updates until the model training process is terminated. It should be noted that the truncation operation of wind resistance change is a key step in controlling the stability of the model training process. By dynamically adjusting the threshold and freezing the parameter updates of some roadway sides in a timely manner, the model can be effectively prevented from overfitting and convergence can be accelerated. The strategy also helps to quickly locate potential abnormal roadways, providing strong support for coal mine safety monitoring.

[0029] This embodiment also provides a computer device applicable to the PINN-based coal mine wind network sparseness wind resistance inversion modeling method, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the PINN-based coal mine wind network sparseness wind resistance inversion modeling method proposed in the above embodiment.

[0030] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0031] This embodiment also provides a storage medium on which a computer program is stored. When executed by a processor, the program implements the PINN-based method for sparse wind resistance inversion modeling of coal mine ventilation networks as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0032] In summary, this invention constructs a sparsed wind resistance inversion model for coal mine ventilation networks by integrating graph neural networks and physical information neural networks. It utilizes minimum spanning tree decomposition and correlation coefficient mechanisms to achieve hard constraint calculation of air volume, effectively ensuring the physical laws of air volume conservation and air pressure balance. By introducing multimodal sensor data and an adaptive sparsity control strategy, the accuracy of wind resistance inversion and the model's generalization ability are significantly improved. This enables accurate identification of abnormal roadways and provides reliable technical support for the safe and efficient operation of coal mine ventilation systems.

[0033] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for inverting wind resistance modeling of sparse coal mine ventilation networks based on PINN, characterized by: include: Obtain the node and roadway topology of the coal mine ventilation network, and construct a graph structure input data including node feature matrix, edge feature matrix and adjacency matrix by combining roadway attribute parameters and multimodal sensor data of air volume, air pressure and temperature. A GNN-PINN hybrid model is established based on the graph structure input data. The topological association features of the wind network are extracted using graph convolutional layers, and the role of co-branch strings in wind volume optimization is enhanced through the edge attention mechanism. In the hybrid model, based on the minimum spanning tree decomposition results, the air volume of the roadway within the spanning tree is directly calculated by the co-branch chord air volume variables through a linear combination of correlation coefficients, forming a hard constraint output that does not participate in gradient updates. Based on the air volume output by the hard constraints and the wind resistance predicted by the model, the wind pressure balance residual and the roadway pressure drop physical relationship residual of each independent loop are calculated, and the fitting residual is constructed by combining the multimodal sensor data to form a soft physical loss term. An adaptive sparsification module is constructed to generate personalized L1 regularization coefficients and dynamic truncation thresholds for each roadway based on roadway type, historical volatility and spatial location. These are used to construct the sparsity loss term and control the truncation operation of wind resistance change. The soft physical loss term and the sparse loss term are combined to form the total loss function, which drives the training of the hybrid model, outputs the inverted wind resistance value of each roadway, and locates abnormal roadways based on the comparison results of wind resistance change and dynamic threshold.

2. The method for sparse wind resistance inversion modeling of coal mine ventilation networks based on PINN as described in claim 1, characterized in that: The specific steps for obtaining the minimum spanning tree decomposition result are as follows: The physical structure of the coal mine ventilation network is abstracted as an undirected connected graph, which consists of a set of nodes and a set of edges. Each element in the set of nodes represents a key connection point in the ventilation network, and each element in the set of edges represents a roadway connecting two nodes. Assign a weight to each edge in the edge set. The value of the weight is the absolute value of the historical air volume data recorded for the roadway under normal and stable ventilation conditions. The minimum spanning tree algorithm is used to process undirected connected graphs to find a subgraph that contains all nodes in the set of nodes. This subgraph satisfies two conditions: the subgraph itself does not contain any cycles and the sum of the weights of all edges in the subgraph is the minimum value among all possible spanning trees. Define the subgraph obtained by the above solution as a spanning tree; The set of all edges in the original edge set that do not belong to the spanning tree is defined as the co-branch chord set.

3. The method for sparse wind resistance inversion modeling of coal mine ventilation networks based on PINN as described in claim 2, characterized in that: The air volume of the generated tree's internal alleyway is directly calculated from the co-branch chord air volume variables through a linear combination of correlation coefficients. The specific steps are as follows: For each co-chord in the set of co-chords, determine the unique basic cycle formed by the co-chord and the edges in the spanning tree; The basic loop is constructed by starting from one endpoint of the co-branch, following the path in the spanning tree to its other endpoint, and then returning to the starting point through the co-branch itself, forming a closed loop. In the basic circuit, all edges except co-chords are identified. All edges belong to the spanning tree and are denoted as spanning tree edges. Define a unified basic loop traversal direction, wherein the basic loop traversal direction is either clockwise or counterclockwise; For each spanning tree edge in the circuit, compare its preset edge direction in the undirected connected graph with the circuit traversal direction to see if they are consistent. If the direction of the spanning tree edge is the same as the direction of the circuit traversal, then set its correlation coefficient with the co-chord to +A; If the direction of the generated tree edge is opposite to the direction of the circuit traversal, then its correlation coefficient with the co-chord is set to -A.

4. The method for sparse wind resistance inversion modeling of coal mine ventilation networks based on PINN as described in claim 3, characterized in that: The specific steps for resolving the wind pressure balance residual are as follows: For each basic loop generated by a co-branch chord, list all the lane edges contained in the loop; For each edge in the loop, calculate its pressure drop based on the edge's wind resistance and the wind volume calculated from the hard constraints output by the model. The expression is: ； in, The inversion drag value of the edge, The airflow value at the edge; Determine the direction coefficient of the edge based on its direction in the loop; If the direction of the edge is the same as the traversal direction of the circuit, the direction coefficient is +B; if the direction is opposite, the direction coefficient is -B. Calculate the algebraic sum of wind pressure for this basic loop. The expression is: ； The summation operation covers all edges in the circuit. Direction coefficient; Will The square of the value is used as the wind pressure balance residual term for this loop.

5. The method for sparse wind resistance inversion modeling of coal mine ventilation networks based on PINN as described in claim 4, characterized in that: The residual physical relationship of the roadway pressure drop is described in the following steps: For each edge in the ventilation network, read the real-time temperature value measured by its associated temperature sensor; Calculate the air density of the region containing the edge using the ideal gas law and the measured temperature. The expression is: ; in, This refers to the local atmospheric pressure. is the gas constant of air; Obtain the initial wind resistance value of this side, and correct it to the equivalent wind resistance under the current environment based on the calculated air density; ; in, This refers to the air density under standard conditions. Using the equivalent wind resistance and the air volume output by hard constraints, the theoretical wind pressure drop on this side is calculated using the following formula: ； If a wind pressure sensor is deployed on this side, the wind pressure drop value measured by it will be obtained; Calculate the pressure drop deviation on this side: when a measured value exists, the deviation is the difference between the theoretical wind pressure drop and the measured wind pressure drop; When no measured value is available, the deviation is the difference between the theoretical wind pressure drop and the wind pressure drop calculated by the model; The square of the deviation is used as the residual term of the physical relationship of the pressure drop on that side.

6. The method for sparse wind resistance inversion modeling of coal mine ventilation networks based on PINN as described in claim 5, characterized in that: The personalized L1 regularization coefficient and dynamic truncation threshold are defined in the following steps: For each lane side, input its lane type identifier, the standard deviation of historical air volume data, and the straight-line distance from the side to the main ventilation fan; Determine the basic sensitivity based on the roadway type identification; The formula for calculating the impact factor of historical air volume fluctuations is as follows: ； in, This represents the maximum standard deviation of historical airflow across all edges in the entire ventilation network. The formula for calculating the spatial distance influence factor is as follows: ; in, This is a preset distance constant; The overall anomaly sensitivity is calculated using the following formula: ； Set the personalized L1 regularization coefficient for this edge. Dynamic truncation threshold ,in, and These are the preset baseline parameters.

7. The method for sparse wind resistance inversion modeling of coal mine ventilation networks based on PINN as described in claim 6, characterized in that: The specific steps for truncating the change in wind resistance are as follows: Set the execution cycle of the truncation operation to every The model training is executed once per iteration. It is a positive integer; If the current training iteration count is When the value is an integer multiple of the specified value, the truncation process is initiated. For each lane, read the change in wind resistance output by the model in the current training iteration and its corresponding dynamic cutoff threshold. Determine whether the absolute value of the change in wind resistance is less than the dynamic cutoff threshold; If the judgment result is yes, then the change in wind resistance on that side will be forcibly set to 0; Mark the edge where the wind resistance change is set to 0 as frozen; In subsequent training iterations, for edges that are frozen, the change in wind resistance remains 0, and they do not participate in gradient calculation and parameter updates until the model training process terminates.

8. A PINN-based coal mine ventilation network sparseness resistance inversion modeling system, based on the PINN-based coal mine ventilation network sparseness resistance inversion modeling method according to any one of claims 1 to 7, characterized in that: include: The module includes a graph data construction module, a model building module, a residual calculation module, a sparsity control module, a loss optimization module, and a result output module. The graph data construction module is used to obtain the node and roadway topology of the coal mine ventilation network, and combine roadway attribute parameters and multi-modal sensor data of air volume, air pressure and temperature to construct graph structure input data including node feature matrix, edge feature matrix and adjacency matrix. The model building module is used to build a GNN-PINN hybrid model based on the graph structure input data, extract the topological correlation features of the wind network using graph convolutional layers, and strengthen the role of co-branch string in wind volume optimization through edge attention mechanism; according to the minimum spanning tree decomposition results, the wind volume of the roadway in the spanning tree is directly calculated by the co-branch string wind volume variables through linear combination of correlation coefficients, forming a hard constraint output that does not participate in gradient update. The residual calculation module is used to calculate the wind pressure balance residual and the roadway pressure drop physical relationship residual of each independent loop based on the wind volume output by hard constraints and the wind resistance predicted by the model, and to construct the fitting residual by combining the multimodal sensor data to form a soft physical loss term. The sparse control module is used to generate personalized L1 regularization coefficients and dynamic truncation thresholds for each roadway based on roadway type, historical volatility and spatial location, to form sparse loss terms and control the truncation operation of wind resistance change. The loss optimization module is used to combine the soft physical loss term and the sparse loss term to form a total loss function, driving the training of the hybrid model; The result output module is used to output the inverted wind resistance value of each roadway and locate abnormal roadways based on the comparison result of wind resistance change and dynamic threshold.

9. A computer device, comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the PINN-based coal mine wind network sparse wind resistance inversion modeling method as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the PINN-based coal mine wind network sparse wind resistance inversion modeling method as described in any one of claims 1 to 7.

Images