A method for simulating ventilation of combined double main holes of a vertical shaft tunnel side adit

Abstract

The present application relates to the cross field of tunnel engineering and computational fluid dynamics, and discloses a kind of vertical shaft tunnel side drift combined double main hole ventilation simulation calculation method without inclination, comprising the following steps: step one, the three-dimensional model of tunnel system is decomposed into first type area and second type area;Step two, for the second type area, by obtaining its high-fidelity simulation data under a variety of boundary conditions, a mathematical proxy model representing the mapping relationship between its input and output boundary state is established;Step three, for specific ventilation simulation task, based on the decomposition result of step one, a global system equation set is constructed and solved.Based on the decomposition of tunnel system into linear flow area and nonlinear node area, a global system equation set with mixed fidelity is finally constructed and solved, which avoids the time-consuming three-dimensional computational fluid dynamics simulation of the entire tunnel system, and realizes the unity of high efficiency and high precision of simulation calculation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of tunnel engineering and computational fluid dynamics, in particular to a method for simulating and calculating ventilation of a non-inclined vertical shaft tunnel combined with a side adit and double main hole. BACKGROUND

[0002] Tunnels are a key component of modern transportation infrastructure, and their operational safety and efficiency are of great importance. In particular, for complex long tunnels, such as double main hole tunnels with non-inclined vertical shafts and side adit auxiliary ventilation, the design and operation control of the internal ventilation system are the core links to ensure daily operating environment and personnel safety evacuation in the event of emergencies such as fires. Therefore, it is of great engineering significance to accurately and efficiently simulate the tunnel ventilation system to predict its airflow organization, temperature and smoke distribution under different working conditions. In the prior art, the numerical simulation method of tunnel ventilation mainly focuses on computational fluid dynamics (CFD) technology. Among them, three-dimensional CFD simulation can provide the highest fidelity description of the details of the tunnel internal flow field by solving the Navier-Stokes equation set, and is recognized as the most accurate analysis method. However, the application of this method is severely restricted by its inherent defects. Due to the large scale of the tunnel structure and the high complexity of the flow, a complete three-dimensional CFD simulation often requires huge computing resources and a long computing period, making it difficult to apply to emergency decision support that requires quick response, or to the design of ventilation schemes that require a large number of parameter optimizations.

[0003] In order to overcome the efficiency bottleneck of three-dimensional CFD simulation, one-dimensional pipe network models are widely used in engineering for simplified calculation. This method abstracts the tunnel system as a network composed of pipe segments and nodes, and the calculation speed is extremely fast. However, its simulation accuracy is heavily dependent on the empirical selection of local resistance coefficients for complex connection areas (such as flow dividing and converging ports). For the areas with significant nonlinear characteristics involved in the present application, such as the connection between the side adit and the main hole, a simple one-dimensional model cannot accurately capture the complex three-dimensional flow details such as separation and vortex in these areas, which will lead to large deviations in the prediction of system air distribution and pollutant diffusion, thereby affecting the reliability of ventilation system design and the effectiveness of emergency control strategies.

[0004] Therefore, the existing tunnel ventilation simulation technology generally faces the contradiction between the calculation accuracy and the efficiency. On the one hand, the high-precision three-dimensional simulation method lacks practicability due to high time cost; on the other hand, the high-efficiency one-dimensional simplified model is difficult to be trusted due to insufficient accuracy, especially under the fire working condition with extremely strict safety requirements. This technical status not only limits the ability of fast and accurate performance evaluation of the complex tunnel ventilation system, but also hinders the development and application of intelligent optimal control strategy based on accurate simulation, so that the current emergency regulation and control of the tunnel ventilation system mainly depends on the preset scheme or artificial experience, and lacks the self-adaptive optimization ability for the emergency situation. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application provides a vertical shaft-free tunnel side drift combined double-main-hole ventilation simulation calculation method, which solves the contradiction between the low calculation efficiency of high-precision three-dimensional simulation and the insufficient calculation accuracy of one-dimensional pipe network model in the existing tunnel ventilation simulation technology, and the problem that the emergency situation cannot be simulated quickly and reliably and the optimal control strategy cannot be formulated.

[0006] To achieve the above object, the present application is implemented by the following technical scheme: a vertical shaft-free tunnel side drift combined double-main-hole ventilation simulation calculation method, comprising the following steps:

[0007] Step one, decompose the three-dimensional model of the tunnel system into a first type of region and a second type of region, the first type of region is a linear flow area with uniform flow field, and the second type of region is a nonlinear node area with complex three-dimensional flow;

[0008] Step two, for the second type of region, high-fidelity simulation data under multiple boundary conditions are obtained to establish a mathematical proxy model representing the mapping relationship between the input and output boundary states;

[0009] Step three, for a specific ventilation simulation task, based on the decomposition result of step one, a global system equation set is constructed and solved, wherein the physical behavior of the first type of region is described by a one-dimensional pipe network model, and the physical behavior of the second type of region is described by the mathematical proxy model established in step two.

[0010] Preferably, the step of topological decomposition of the tunnel system specifically comprises: abstracting each nonlinear node area as a node of a network topology graph, and abstracting each linear flow area connecting the nonlinear node areas as an edge connecting the nodes, thereby generating a network topology graph representing the physical connection relationship of the tunnel system.

[0011] Preferably, the step of constructing the network topology graph further comprises, for each edge of the network topology graph, assigning physical property parameters required by the corresponding linear flow zone in the one-dimensional pipe network model to the edge, the physical property parameters including pipe segment length, hydraulic diameter and friction coefficient of the linear flow zone.

[0012] Preferably, the step of constructing the mathematical surrogate model comprises, obtaining the high-fidelity simulation data by performing multiple three-dimensional computational fluid dynamics simulations, and training a machine learning algorithm based on the high-fidelity simulation data to obtain the mathematical surrogate model.

[0013] Preferably, the mathematical surrogate model outputs a prediction confidence indicator for characterizing the confidence of the prediction result output by the mathematical surrogate model according to the input boundary conditions.

[0014] Preferably, the step of constructing and solving the global system of equations further comprises an adaptive correction step triggered based on the prediction confidence indicator output by the mathematical surrogate model.

[0015] Preferably, the adaptive correction step comprises, in the process of solving the global system of equations, monitoring the prediction confidence indicator in real time, and when the prediction confidence indicator is lower than a preset threshold, performing a three-dimensional computational fluid dynamics simulation in real time for the corresponding nonlinear node zone to obtain a high-precision online simulation result, and using the online simulation result to update the mathematical surrogate model online.

[0016] Preferably, the method further comprises constructing a topology state graph, wherein the nodes of the topology state graph represent macroscopic states of the tunnel system, and the edges of the topology state graph represent control operations that can cause transitions between the macroscopic states.

[0017] Preferably, the method further comprises, for a preset control target, for at least part of the edges in the topology state graph, determining the system evolution result caused by the control operation represented by the edge by constructing and solving the global system of equations, and calculating the weight of the edge according to the system evolution result.

[0018] Preferably, the method further comprises, on the topology state graph, based on the weights of the edges, using a graph search algorithm to search for a path with optimal cumulative weight from a starting state node to a target state node, and outputting the control operation sequence corresponding to the optimal path.

[0019] The application provides a vertical shaft tunnel side adit combined double-main-hole ventilation simulation calculation method. The application has the following beneficial effects:

[0020] 1. The present application decomposes the tunnel system into linear flow regions and nonlinear node regions, and uses one-dimensional pipe network models and pre-trained mathematical agent models to describe them respectively, finally constructs and solves a global system equation set with mixed fidelity, which greatly improves the computational efficiency of the whole system ventilation simulation while ensuring the accurate capture of the physical behavior of key complex regions. This method avoids the time-consuming three-dimensional computational fluid dynamics simulation of the entire tunnel system, and realizes the unity of high efficiency and high precision in simulation calculation.

[0021] 2. The present application introduces an adaptive correction step based on the confidence index of the mathematical agent model prediction in the process of solving the global system equation set, which realizes the adaptability and high reliability of the simulation calculation. This method can monitor the prediction reliability of the agent model under unknown working conditions in real time, and when its credibility is insufficient, it can call high-precision simulation for targeted data supplement and model update on demand and online, so as to ensure that the entire simulation framework can still output stable and reliable calculation results even when facing rare or extreme working conditions that are not covered by the training data.

[0022] 3. The present application builds a topological state atlas representing the macroscopic state of the system and the control operation, and uses the aforementioned efficient and accurate simulation calculation capability to dynamically assign costs to the control operations in the atlas based on the physical evolution results, and finally combines graph search algorithm to deduce the optimal control path, which upgrades the traditional simulation calculation tool to an intelligent decision support system. This method can automatically and quickly generate the optimal control operation sequence with quantitative evaluation for specific control targets, realizing the intelligentization and optimization of the ventilation control strategy. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 The flowchart of the method of the present application. DETAILED DESCRIPTION

[0024] The technical solutions in the embodiments of the present application will be described below in conjunction with the drawings of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0025] Please refer to the drawings of the present application Figure 1An embodiment of the present invention provides a ventilation simulation calculation method for a side guide tunnel combined with dual main tunnels in a non-slanted vertical shaft tunnel. The method decomposes the tunnel system into a linear flow area and a nonlinear node area, and describes them respectively using a one-dimensional pipe network model and a pre-trained mathematical proxy model. Finally, a mixed-fidelity global system equation group is constructed and solved. While ensuring the accurate capture of the physical behavior of key complex areas, the computational efficiency of the ventilation simulation of the entire system is greatly improved.

[0026] like Figure 1 As shown, the ventilation simulation calculation method for the side guide tunnel of a non-slanted vertical shaft tunnel combined with dual main tunnels may include the following steps: Step 1, decomposing the three-dimensional model of the tunnel system into a first type of area and a second type of area, wherein the first type of area is a linear flow area with uniform flow field, and the second type of area is a nonlinear node area with complex three-dimensional flow;

[0027] Step 2: For the second type of region, by obtaining high-fidelity simulation data under various boundary conditions, a mathematical proxy model is established to characterize the mapping relationship between its input and output boundary states;

[0028] Step 3: For specific ventilation simulation tasks, based on the decomposition results of step 1, construct and solve the global system equations. The physical behavior of the first type of area is described by the one-dimensional pipe network model, and the physical behavior of the second type of area is described by the mathematical proxy model established in step 2.

[0029] The method begins with a detailed topological decomposition of the tunnel system consisting of a non-slanted vertical shaft tunnel with side guides and dual main tunnels. This step serves as a foundation for the subsequent construction of a hybrid-fidelity simulation model that balances computational accuracy and efficiency. In tunnel engineering, the complexity of fluid motion varies significantly across different regions, making it difficult to balance accuracy and cost using a unified computational model. Therefore, this step begins by identifying and classifying the tunnel system based on its physical characteristics.

[0030] Specifically, this segmentation process is based on an analysis of the complete 3D geometry of the tunnel system and its inherent fluid dynamics. This analysis is based on factors such as geometric curvature, cross-sectional change rates, the presence of diverging and converging flows, and the presence of interference from external power sources (such as fans). Based on these factors, the entire tunnel system is physically divided into two distinct regions:

[0031] The first type of region, namely linear flow zone. This type of region is typically characterized by its well-developed and uniform internal flow field, with the flow direction being essentially parallel to the tunnel axis, and the three-dimensional effect being insignificant. In this type of region, the energy loss of the fluid is mainly dominated by the wall friction along the path, and its physical behavior can be accurately described by one-dimensional macroscopic parameters such as the average flow velocity and the average pressure of the section. In the specific scenario of the present invention, the linear flow zone generally refers to the long straight tunnel section far away from the complex structures such as intersections, connections, etc., such as the middle straight section of the double main hole, and the straight section of the side guide hole.

[0032] The second type of region, namely nonlinear node zone. This type of region is in sharp contrast to the linear flow zone, and there are strong and non-negligible complex three-dimensional flow phenomena inside. These phenomena include but are not limited to flow separation and reattachment, secondary flow formation, strong vortex motion, and significant energy and momentum exchange. These complex physical processes make simple one-dimensional models completely invalid. In the tunnel structure targeted by the present invention, the nonlinear node zone explicitly includes the connection between the side guide hole and the main hole, the T-shaped or cross-shaped connection between the transverse connecting channel and the main hole, and the influence section of the jet fan group set up to meet the ventilation requirements. Under accident conditions such as fire, the fire source itself and its nearby hot buoyancy dominant region should also be considered as a dynamically formed nonlinear node zone.

[0033] After completing the physical region division of the tunnel system, in order to facilitate subsequent numerical solution at the system level, this step also includes a key topological abstraction process. This process converts the aforementioned physical division results into a structured mathematical expression.

[0034] Specifically, this topological abstraction step abstracts each identified nonlinear node zone into a node in a network topology graph. These nodes in the graph represent "junctures" that have complex physical behavior and need to be analyzed in detail.

[0035] Correspondingly, each linear flow zone connected between two nonlinear node zones is abstracted into an edge connecting the corresponding two nodes in the graph. These edges in the graph represent the "channels" for the macroscopic transport of fluid.

[0036] Through the above abstraction process, a complex, continuous three-dimensional tunnel entity is converted and simplified into a discrete network topology graph composed of nodes and edges. Preferably, the network topology graph can be expressed in mathematical form as G = (V, E), where V is the node set consisting of all nonlinear node zones, and E is the edge set consisting of all linear flow zones. This transformation greatly simplifies the complexity of the problem, making it possible to build and solve the global equation system subsequently.

[0037] It needs to be further explained that the generated network topology graph not only contains the connection relationship, but also must carry the physical information required for subsequent calculation. Therefore, this step also includes assigning physical attribute parameters to the network topology graph.

[0038] Specifically, for each edge in the network topology graph, i.e. each linear flow region, a set of physical attribute parameters necessary in the one-dimensional pipe network model needs to be calculated and assigned. These parameters are the direct basis for describing the physical behavior of the region in the subsequent steps. In this embodiment, the physical attribute parameters at least include:

[0039] Pipe segment length: the actual length of the linear flow region along its center line.

[0040] Hydraulic diameter: an equivalent diameter calculated according to the cross-sectional geometry of the linear flow region, used for one-dimensional flow loss calculation.

[0041] Friction coefficient along the way: a dimensionless coefficient representing the frictional resistance of fluid flowing through the inner wall of the linear flow region, which is related to the roughness of the tunnel wall and the Reynolds number.

[0042] Step two of the method of the present application is to build a mathematical proxy model for each type of nonlinear node region identified in step one, which can accurately reproduce its complex physical behavior while being extremely lightweight in calculation. The setting of this step is the key to realizing the unity of high efficiency and high precision in the present application. It is not practical to frequently call three-dimensional computational fluid dynamics (CFD) models in system-level simulation, and the mathematical proxy model is exactly a high-fidelity "digital twin" of the CFD model, which is integrated for use in subsequent calculations.

[0043] The specific implementation of this step can be further divided into a high-fidelity simulation data acquisition process and a data-based model training process.

[0044] First, high-fidelity simulation data acquisition. For a specific type of nonlinear node region, such as a side tunnel-main tunnel T-shaped connection port with three ports, the internal mapping relationship between its input and output boundary states needs to be systematically explored.

[0045] The first step of this process is parameterization definition. Define the fluid state parameters of all physical ports of the nonlinear node region, such as static pressure , flow rate , temperature and key component concentration , as a boundary state vector. Part of the parameters are used as input vector , and the other part as output vector Next, for efficiently exploring the entire working parameter space of the nonlinear node region, preferably, a Design of Experiments (DoE) method, such as Latin Hypercube Sampling (LHS), can be adopted to generate a set of representative and uniformly distributed input boundary condition sample points within a reasonable range of its input parameters For each generated input sample point , the method will perform a complete and high-precision three-dimensional computational fluid dynamics simulation. By solving the Navier-Stokes equations, the stable three-dimensional flow field distribution inside the nonlinear node region under the input condition can be obtained, and the corresponding output boundary state vector is extracted therefrom. After the calculation of all sample points is completed, a high-fidelity simulation database that can fully characterize the physical properties of the nonlinear node region is obtained, which is composed of a large number of "input-output" data pairs .

[0046] Then, the process of model training based on the obtained data. The goal of this process is to use machine learning algorithms to learn and fit a continuous and generalizable mathematical function, i.e., a mathematical surrogate model , from discrete high-fidelity data points , so that .

[0047] Preferably, in this embodiment, Gaussian Process Regression (GPR) algorithm can be used to construct the mathematical surrogate model. The reason for choosing Gaussian Process Regression is its unique non-parametric modeling capability based on Bayesian theory. It not only provides prediction of output results, but also quantitatively evaluates the uncertainty of prediction, which is crucial for the subsequent adaptive correction of the invention.

[0048] Specifically, when a new input boundary condition is given outside the database, the trained Gaussian Process Regression model can output two types of information simultaneously:

[0049] First, the prediction result used to represent the output boundary state, i.e., the prediction mean . This prediction result is the best estimate of the true physical output , which can be directly used for subsequent system-level simulation calculations.

[0050] Second, the prediction confidence index used to represent the credibility of the prediction result, i.e., the prediction variance . This index quantifies the "confidence" of the model itself on the accuracy of this prediction. Generally speaking, when the input When the input is close to the sample points in the training database, the prediction variance is small, indicating that the prediction result is highly reliable; on the contrary, when the input is far away from all known data points, in the "cognitive blind area" of the model, the prediction variance will significantly increase, thus sending a warning signal that the reliability of the current prediction result is low.

[0051] Through the above process, each type of nonlinear node region is equipped with a dedicated and well-trained mathematical agent model. These models provide prediction accuracy similar to three-dimensional CFD simulation at a very low computational cost, and creatively add the ability to diagnose their prediction reliability, thus providing core technical support for building a robust and efficient adaptive hybrid fidelity solution framework in step three.

[0052] Step three of the method is the core calculation link for performing specific ventilation simulation tasks. Instead of using a single scale model for solving, this step constructs and solves an innovative, multi-fidelity coupled global system equation set based on the topological decomposition results of step one and the mathematical model library constructed in step two. This design aims to organically integrate the computational efficiency of one-dimensional models and the physical precision of three-dimensional models.

[0053] First, the construction process of the global system equation set. For a given ventilation simulation task, such as smoke diffusion analysis in a specific fire scenario, this method automatically assembles a large nonlinear equation set describing the steady-state or transient behavior of the entire tunnel system based on the network topology graph generated in step one, which carries physical properties. The "hybrid fidelity" feature of this equation set is reflected in its use of different mathematical description methods for different regions:

[0054] For each edge in the network topology graph, i.e., each tunnel segment divided into a linear flow region, the internal physical behavior is described by a classical one-dimensional pipe network model. Specifically, the relationship between pressure drop and flow can be characterized by the Darcy-Weisbach formula:

[0055]

[0056] where is the pressure difference between the two ends of the pipe segment, and the pipe segment length , hydraulic diameter , and friction coefficient are physical property parameters assigned to the edge in step one. For each node in the network topology graph, i.e., each complex junction identified as a nonlinear node region, its physical behavior is described by the pre-trained mathematical agent model ​​The model is described in a functional form The nonlinear exchange relationships between the pressure, flow rate, energy and component concentration of all ports of the node are given, which replaces the calculation results that must be obtained through expensive three-dimensional CFD simulation. In addition, in order to ensure the physical conservation of the entire system, the mass conservation, momentum conservation and energy conservation constraint equations must also be established and satisfied on the interface between each nonlinear node region and the linear flow region connected thereto. Combining all the equations for the edges, nodes and interfaces, a complete global system equation set describing the behavior of the entire tunnel system is formed . Among them, is a global state vector containing all unknown quantities to be solved (for example, the pressure of each node and the flow rate of each pipe section). Preferably, a robust numerical iterative method such as the Newton-Raphson method can be used to solve the large nonlinear equation set.

[0057] However, one of the core innovations of this step is not only the hybrid fidelity modeling described above, but also the inclusion of an intelligent adaptive correction step. The setting of this step aims to solve the problem of decreased prediction accuracy of the mathematical agent model for unknown working conditions outside the coverage of the training database, thereby ensuring the robustness and reliability of the entire simulation process.

[0058] This adaptive correction step is deeply coupled with the iterative solution process of the global system equation set, and its triggering is completely based on the prediction confidence index output by the mathematical agent model of step two. The specific implementation process is as follows:

[0059] First, in each iteration of solving the global system equation set, when the system calls the mathematical agent model of any nonlinear node region to calculate its output , the system not only obtains its prediction result , but also synchronously obtains its prediction confidence index, such as the prediction variance given by the Gaussian process regression model. This is a real-time monitoring process.

[0060] Next, the system compares the prediction confidence index with a preset, user-defined threshold. When it is found that the prediction confidence index of a certain nonlinear node region is lower than the preset threshold, for example threshold, this constitutes a clear trigger signal. This signal indicates that the current input working condition is in the weak area of the agent model's cognitive ability, and the reliability of its prediction result is not sufficient to support subsequent calculations.

[0061] Once the correction step is triggered, the system will immediately perform an online update procedure. The procedure will first pause the iterative solution of the main program. Then, for the nonlinear node region that raised the alarm, with its current, low-confidence prediction-inducing input boundary conditions a complete three-dimensional computational fluid dynamics simulation is performed online, on-demand, for the input. This simulation will obtain a high-precision online simulation result for the current unknown operating condition . Subsequently, this newly obtained high-fidelity data point will be used as a new sample to online update or retrain the mathematical surrogate model, thus improving its knowledge level and prediction accuracy in this unknown region.

[0062] After the online update of the mathematical surrogate model, the iterative solution process of the main program will resume from the pause and continue using this "target-enhanced", more reliable model until the entire global system of equations converges.

[0063] Through this closed-loop adaptive correction mechanism of monitoring, triggering, online simulation and model updating, the method of the present invention ensures the stability of the calculation process and the accuracy of the final result. It can intelligently identify and compensate for the cognitive shortcomings of the surrogate model, thus maximizing the advantages of hybrid fidelity simulation.

[0064] First, this step needs to build a topological state atlas with a higher dimension than the physical network topology graph in step one. This atlas is a decision graph, whose nodes represent the macroscopic states of the entire tunnel system. A macroscopic state is defined by two parts of information: one is the state of physical events (such as the specific location of the fire source in the tunnel, the burning power, etc.), the other is the state of controllable devices (such as the opening / closing / regulating state of each jet fan, axial flow fan, damper valve, etc.). The edges of the atlas represent an atomic control operation that can cause a transition between macroscopic states, such as "turn on the No. 1 jet fan group" or "adjust the speed of the No. 2 main hole exhaust fan to 80%".

[0065] After receiving a preset control target (for example, in a certain fire scenario, to reduce the smoke concentration in the evacuation passage to below the safety threshold within a specified time with the lowest energy consumption), the method converts this complex decision-making problem into a shortest path search problem on the topological state atlas. The core of this process is to dynamically and physically meaningfully weight the edges in the atlas.

[0066] Specifically, for a certain edge in the topology state map to be evaluated (i.e. a candidate control operation), the system will invoke and execute the aforementioned step three hybrid fidelity ventilation simulation method completely, with the system state represented by the starting node of the edge as the initial condition. By constructing and solving the global system equation set, the system can efficiently predict the next stable state that the tunnel system will evolve to after the control operation is performed, i.e. obtain the system evolution result caused by the operation.

[0067] Subsequently, the system will quantitatively evaluate the system evolution result according to a pre-set cost function (which comprehensively considers various indicators of interest to the control target, such as personnel safety, smoke exhaust efficiency, control time, equipment energy consumption, etc.), and calculate a scalar value. This value is assigned as the weight of the edge.

[0068] After assigning weights to the relevant edges in the topology state map, the method preferably adopts an efficient graph search algorithm such as A* (A-star) to search for a path with the optimal cumulative weight from the starting node (i.e. the initial state of the accident) to the target node (i.e. the state satisfying the control target) on the graph that has been dynamically weighted.

[0069] This optimal path itself clearly indicates a series of consecutive control steps with the minimum cost. Finally, the method outputs the control operation sequence corresponding to the optimal path, providing clear, reliable and quantitatively evaluated intelligent decision support for tunnel operators.

[0070] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and variations can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A ventilation simulation calculation method for a non-slanted vertical shaft tunnel with a side guide tunnel and dual main tunnels, characterized in that: The following steps are involved: Step 1: Decompose the three-dimensional model of the tunnel system into a first type of region and a second type of region, wherein the first type of region is a linear flow region with uniform flow field, and the second type of region is a nonlinear node region with complex three-dimensional flow; Step 2: For the second type of region, by obtaining high-fidelity simulation data under various boundary conditions, a mathematical proxy model is established to characterize the mapping relationship between its input and output boundary states; Step 3: For a specific ventilation simulation task, based on the decomposition results of step 1, construct and solve a global system equation set, wherein the physical behavior of the first type of area is described by a one-dimensional pipe network model, and the physical behavior of the second type of area is described by the mathematical proxy model established in step 2; The step of performing topological decomposition on the tunnel system specifically includes abstracting each of the nonlinear node regions into a node of a network topology graph, and abstracting each of the linear flow regions connecting the nonlinear node regions into an edge connecting the nodes, thereby generating a network topology graph representing the physical connection relationship of the tunnel system; The step of drawing the network topology diagram further includes assigning physical property parameters required by the linear flow area corresponding to each edge of the network topology diagram in a one-dimensional pipe network model, wherein the physical property parameters include the pipe segment length, hydraulic diameter, and friction coefficient along the linear flow area; The steps of the mathematical proxy model are specifically as follows: performing multiple three-dimensional computational fluid dynamics simulations to obtain the high-fidelity simulation data, and using a machine learning algorithm to train based on the high-fidelity simulation data to obtain the mathematical proxy model.

2. A ventilation simulation calculation method for a non-slanted vertical shaft tunnel side guide tunnel combined with dual main tunnels according to claim 1, characterized in that: The step of constructing and solving the global system equations further includes an adaptive correction step, which is triggered based on the prediction confidence index output by the mathematical agent model.

3. A ventilation simulation calculation method for a non-slanted vertical shaft tunnel side guide tunnel combined with dual main tunnels according to claim 2, characterized in that: The adaptive correction step specifically includes: monitoring the prediction confidence index in real time during the process of solving the global system equations, and when the prediction confidence index is lower than a preset threshold, performing a three-dimensional computational fluid dynamics simulation online for the corresponding nonlinear node area to obtain a high-precision online simulation result, and using the online simulation result to update the mathematical agent model online.

4. The method for simulating ventilation of a non-slanted vertical shaft tunnel with a side guide tunnel and dual main tunnels according to claim 1 is characterized in that: The method further includes: constructing a topological state graph, wherein the nodes of the topological state graph represent macroscopic states of the tunnel system, and the edges of the topological state graph represent control operations that can cause transitions between the macroscopic states.

5. The method for simulating ventilation of a non-slanted vertical shaft tunnel with a side guide tunnel and dual main tunnels according to claim 1 is characterized in that: The method also includes: for a preset control target, for at least part of the edges in the topological state graph, constructing and solving the global system equations to determine the system evolution result caused by executing the control operation represented by the edge, and calculating the weight of the edge based on the system evolution result.

6. The method for simulating ventilation of a non-slanted vertical shaft tunnel with a side guide tunnel and dual main tunnels according to claim 1 is characterized in that: The method further includes: searching for a path with the optimal cumulative weight from the starting state node to the target state node on the topological state graph based on the weight of the edge using a graph search algorithm, and outputting a control operation sequence corresponding to the optimal path.

Images