A comprehensive pipe gallery fire positioning method based on modal identification and Bayesian algorithm

Abstract

The application discloses a comprehensive pipe gallery fire positioning method based on modal identification and a Bayesian algorithm, first, for the spatial scene, the comprehensive pipe gallery disaster situation is determined, and a comprehensive pipe gallery fire numerical calculation model is established based on a computational fluid dynamics method; second, a physical field modal reduction model is constructed based on time slice data, and nonlinear modal identification of physical information such as temperature, carbon dioxide and carbon monoxide in the comprehensive pipe gallery during a fire is realized; third, a neural network driven by physical information is used to perform nonlinear modal feature identification on the features of the physical information such as temperature, carbon dioxide and carbon monoxide; and finally, a Bayesian inference algorithm method is used for rapid positioning of the fire position. The application integrates machine learning theory, improves the accuracy, interpretability and timeliness of the model, and has important practical application value for rapid prediction and disaster source positioning of the comprehensive pipe gallery, especially in emergency situations, can quickly respond, and provides support for disaster management.

Description

Technical Field

[0001] This invention relates to a comprehensive method for locating fires in utility tunnels based on modality recognition and Bayesian algorithms, belonging to the field of fire location technology. Background Technology

[0002] Integrated utility tunnels are a type of modern urban infrastructure primarily used to centrally house and protect various underground pipelines, such as power, communication, gas, water supply, and drainage systems. This design not only optimizes the use of urban space and reduces maintenance costs but also significantly improves the efficiency and safety of urban operations. However, underground integrated utility tunnels have a long, narrow, and enclosed spatial structure with complex internal pipeline arrangements, resulting in a high fire hazard. Fires in these tunnels can cause significant damage, with fires in cable compartments being the most severe. Therefore, it is necessary to study the fire problems of cable compartments within integrated utility tunnels.

[0003] Currently, fire parameter identification and state construction mainly rely on traditional methods or machine learning methods. Traditional methods have significant limitations and struggle to handle complex scenarios, while machine learning requires sufficient training samples, is highly dependent on training data, and has poor generalization ability. In real-world disaster scenarios, sensor deployment areas are limited, making it impossible to obtain sufficient data samples. Furthermore, models that rely solely on data-driven approaches may yield theoretical results that violate physical theorems, leading to significant errors in the analysis of disaster scenarios that are difficult to observe directly. However, disaster data typically contains certain physical laws, empirical rules, or other boundary information. Current machine learning methods do not analyze and utilize the inherent patterns and underlying physical models within the limited amount of disaster information data. Therefore, a new method is needed to predict integrated utility tunnel disaster scenarios and quickly locate the disaster source. Summary of the Invention

[0004] The technical problem to be solved by this invention is to provide a method for locating fires in integrated utility tunnels based on modal recognition and Bayesian algorithms, which can effectively predict the spread of the physical field during a fire in an integrated utility tunnel, invert the location of the fire ignition point based on Bayesian theory and MCMC algorithm, and use sensor parameters to quickly locate the fire source.

[0005] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0006] A method for locating fires in utility tunnels based on modality recognition and Bayesian algorithms includes the following steps:

[0007] Step 1: Construct a CFD numerical model of a fire scenario in a utility tunnel and solve the model to obtain the physical field information of the utility tunnel as it changes over time.

[0008] Step 2: Based on the physical field information of the integrated utility tunnel that changes over time obtained in Step 1, construct time slice data, and use a nonlinear sparse identification algorithm to extract the physical field mode reduction model from the time slice data.

[0009] Step 3: Construct a physical information neural network model that takes the location of the fire source and the corresponding temperature information as input and the predicted temperature field information inside the integrated utility tunnel as output. Obtain the first type of initial training set, the second type of training set and the third type of training set according to the physical field mode reduction model. Use the first type of initial training set to pre-train the physical information neural network model to obtain the pre-trained model. Further train the pre-trained model using the second and third type of training sets to obtain the trained model.

[0010] Step 4: Reconstruct the temperature field within the integrated utility tunnel using the trained model, and quickly locate the fire source based on the reconstructed temperature field and Bayesian inference algorithm.

[0011] As a preferred embodiment of the present invention, the specific process of step 1 is as follows:

[0012] Based on different fire source locations, fire intensity, and ventilation conditions inside the utility tunnel, a CFD numerical model of a fire scenario in the utility tunnel is constructed using computational fluid dynamics (CFD) technology. The model is then solved using a semi-implicit algorithm based on the pressure coupling equations to obtain the physical field information of the utility tunnel that changes over time, including data on temperature, carbon monoxide concentration, carbon dioxide concentration, and oxygen concentration during a fire.

[0013] As a preferred embodiment of the present invention, in step 2, based on the physical field information of the integrated utility tunnel that changes over time obtained in step 1, the slicing time is set to 0.1s, and time slice data of fire source location, temperature, carbon monoxide concentration, carbon dioxide concentration and oxygen concentration in the physical field information are extracted every 0.1s.

[0014] The formula for the physical field modal order reduction model is as follows:

[0015]

[0016] In the formula, X is the state variable, Θ(X) is the feature library function matrix, and Θ1(X), Θ2(X), Θ... n (X) are all feature library functions, and Ξ is a sparse regression coefficient matrix used to describe the physical field modal reduction model, ∈1(X), ∈2(X), ∈ n (X) are all sparse regression coefficients.

[0017] As a preferred embodiment of the present invention, the specific process of step 3 is as follows:

[0018] A physical information neural network model is constructed, which takes the location of the fire source and the corresponding temperature information as input and the predicted internal temperature field information of the integrated utility tunnel as output.

[0019] A point is randomly selected from other locations in the integrated utility tunnel except for the boundary area as the first sampling point. The location of the first sampling point and the corresponding physical field information are obtained as a training data in the first type of initial training set. The process of randomly selecting the first sampling point and obtaining the corresponding training data is repeated multiple times to generate the first type of initial training set.

[0020] A point is randomly selected in the boundary area of ​​the integrated utility tunnel as the second sampling point. The location of the second sampling point and the corresponding physical field information are obtained as a training data in the second type of training set. The process of randomly selecting the second sampling point and obtaining the corresponding training data is repeated multiple times to generate the second type of training set.

[0021] Based on the heat conduction equation and the boundary of the integrated utility tunnel, a first-type loss function is constructed. The physical information neural network model is then pre-trained using the first-type initial training set and the first-type loss function to obtain model parameters that minimize the first-type loss function, which are then used as the parameters of the pre-trained model. The heat conduction equation is as follows:

[0022]

[0023] In the formula, k is the thermal conductivity coefficient, T is the temperature, φ(x,y) is the heat source term, which represents the internal heat source or heat generated per unit volume, and (x,y) is the position coordinate;

[0024] The formula for the first type of loss function is as follows:

[0025]

[0026] In the formula, Let w represent the first type of loss function. pde w bc For hyperparameters, L pde For the loss of the internal partial differential equation, L bc For boundary condition loss, P pde For the first type of initial training set, T θ The output temperature field is θ, where θ is the model parameter and φ is the model parameter. rated Let P be the heat source intensity distribution function. bc For the second type of training set, T0 is the temperature at the vent location or the ambient temperature, and n is the direction of the boundary normal.

[0027] A second type of loss function is constructed based on the heat conduction equation and the boundary of the integrated utility tunnel. The pre-trained model is then trained again using the second type of training set and the second type of loss function to obtain the model parameters that minimize the second type of loss function, which are then used as the parameters of the trained model. The formula for the second type of loss function is as follows:

[0028]

[0029] In the formula, Let w represent the second type of loss function. data For hyperparameters, L data For the loss term based on the temperature sensor monitoring value, T obd P is the temperature sensor reading. data This is the third type of training set; the third type of training set consists of training data composed of the temperature monitored by temperature sensors deployed in the integrated utility tunnel and the location coordinates of the temperature sensors.

[0030] As a preferred embodiment of the present invention, the specific process of step 4 is as follows:

[0031] Step 4.1: Use the temperature, corresponding time, and location coordinates of the temperature sensors installed in the integrated utility tunnel as inputs to the Bayesian inference algorithm.

[0032] Step 4.2: Assuming the fire source is located at the exact center of the utility tunnel, reconstruct the temperature field within the utility tunnel based on the model trained in Step 3.

[0033] Step 4.3: Calculate the likelihood probability of the temperature monitored by each temperature sensor under the temperature field reconstructed in step 4.2;

[0034] Step 4.4: Based on the likelihood probability and the temperature monitored by the temperature sensor, the location of the fire source is calculated using the Markov chain Monte Carlo method.

[0035] Step 4.5: Determine whether the fire source location calculated in Step 4.4 is consistent with the fire source location set in Step 4.2. If so, use the fire source location calculated in Step 4.4 as the final fire source location output; otherwise, proceed to Step 4.6.

[0036] Step 4.6: Replace the fire source location set in step 4.2 with the fire source location calculated in step 4.4, and repeat steps 4.2 to 4.5.

[0037] A computer device includes a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the processor executes the computer program to implement the steps of the integrated utility tunnel fire location method based on modal recognition and Bayesian algorithm.

[0038] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the integrated utility tunnel fire location method based on modal recognition and Bayesian algorithm.

[0039] Compared with the prior art, the present invention, employing the above technical solution, has the following technical effects:

[0040] 1. This invention analyzes and utilizes the inherent patterns and underlying physical models of a limited amount of disaster information data, overcoming the limitations of traditional data-driven machine learning data modeling and breaking the black-box modeling of neural networks, thus greatly improving the accuracy and interpretability of the model. By analyzing models of physical information such as temperature, carbon dioxide, and carbon monoxide, as well as their corresponding feature parameters, the invention uses Bayesian theory and the MCMC algorithm to predict the location of fire ignition points within utility tunnels, achieving rapid response in the event of a fire.

[0041] 2. This invention fully considers the physical characteristics of the enclosed scenario of integrated utility tunnels, using known boundary conditions as modal constraints to solve the reduced-order modal model of the physical field. Furthermore, it integrates with machine learning methods, providing significant practical value for the rapid prediction and source location of fires in integrated utility tunnels, especially enabling rapid response in emergency situations and supporting disaster management. Attached Figure Description

[0042] Figure 1 This is a flowchart of a comprehensive utility tunnel fire location method based on modal recognition and Bayesian algorithm according to the present invention.

[0043] Figure 2 This is a flowchart of the CFD numerical model solution process;

[0044] Figure 3 This is a flowchart of a physical field modal order reduction model built based on time-slice data;

[0045] Figure 4 This is a diagram of the physical information neural network model structure;

[0046] Figure 5 This is a flowchart of nonlinear modal feature recognition based on a physical information neural network model;

[0047] Figure 6 This is a flowchart of Bayesian inference;

[0048] Figure 7 This is a numerical simulation diagram of a fire in a utility tunnel, where (a) represents the temperature of the tunnel and (b) represents the smoke from the tunnel.

[0049] Figure 8 The accuracy of fire source tracing in each model;

[0050] Figure 9 It is a temperature field simulation of different fire source locations in the integrated utility tunnel. Detailed Implementation

[0051] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0052] The integrated utility tunnel fire location method based on modality recognition and Bayesian algorithm proposed in this invention mainly consists of four parts, the specific ideas of which are as follows: Figure 1 As shown.

[0053] 1. CFD numerical model of disaster scenarios in integrated utility tunnels

[0054] Computational fluid dynamics (CFD) is the computation and study of fluids. It has been gradually developed and improved based on multiple disciplines such as fluid mechanics theory and computer science. Its main research goal is to realize the study of the laws of fluid mechanics in a more scientific and efficient way, focusing on the discretization and solution analysis process of integral and differential terms in fluid control equations.

[0055] In this invention, the construction of the integrated utility tunnel fire model mainly utilizes general-purpose CFD software to build various scenarios of the integrated utility tunnel. The calculation method employs numerical methods to solve a set of Navier-Stokes equations for low-speed flow. During the calculation process, based on thermal drive, the heat transfer process, smoke flow, and fire spread in the fire are analyzed. The software divides the space of the analysis object into multiple extremely small three-dimensional grids or computational units, solving for the parameters of each unit, including temperature, gas density, flow velocity, pressure, and component concentration. In the calculation, fluid dynamics partial differential equations such as mass conservation, momentum conservation, and energy conservation are used to approximate finite difference. By dividing the grid, the finite volume method is used to calculate thermal radiation and analyze turbulence phenomena in the fluid, thereby tracking and predicting the movement of gas in the fire, and combining the parameter characteristics of various materials in the constructed model to extrapolate the growth and spread of the fire. In this software, a semi-implicit algorithm based on pressure coupling equations is used for calculation. This algorithm is a pressure correction method, mainly following the idea of ​​"pressure prediction-pressure correction," solving the pressure field in the discretized fluid element grid, and further calculating and solving the momentum equation. The research problem involves diffusion and transport between different gas components, and the process of gas injection from the pipe into the pipe gallery exhibits certain turbulent characteristics. The corresponding solution flow diagram is shown below. Figure 2 As shown.

[0056] Through the above simulations, the changes in temperature, carbon monoxide, carbon dioxide, and oxygen concentrations under various conditions during a fire in a utility tunnel can be obtained. The accuracy and effectiveness of the model used in this invention were verified through a full-scale fire test in a utility tunnel. The test simulated numerous fire scenarios in the cable compartment of the utility tunnel caused by cable joint fires, short circuits, and other situations, recreating the real-world fire scenarios in utility tunnels to the greatest extent possible.

[0057] 2. Physical field modal order reduction model constructed based on time slice data

[0058] Based on the aforementioned model, this section first extracts the physical field information within the integrated utility tunnel that changes over time, including data such as temperature, carbon dioxide, and carbon monoxide, to construct time-slice data. Then, a nonlinear sparse identification algorithm is used to extract the nonlinear physical field model from the time-slice data, thereby achieving the result of modal order reduction of the physical field.

[0059] Nonlinear sparse identification algorithms are efficient methods for extracting reduced-order dynamic models from time-slice data (including spatial, temporal, and physical field data). Specifically, the behavior of a dynamic system can typically be formalized using differential equations, where the rate of change of the system's state over time is described by an unknown function f. A feature library is constructed from the system state data, containing various functions (such as polynomials and trigonometric functions) that may influence the system's dynamics. This library aims to cover all potential dynamic behaviors the system might exhibit. It is assumed that the actual system dynamics can be accurately described by a few functions in the feature library. Therefore, the goal is to find a sparse vector that best represents these key features while ensuring that the dynamic model constructed from the feature library approximates the dynamics of the real system as closely as possible.

[0060] The model process is as follows: Figure 3 As shown, the process begins with data collection and preprocessing to ensure data quality and consistency. Next, a rich feature library is built based on this data, containing polynomials, cross terms, and trigonometric functions of the original variables. Through sparse regression analysis, such as LASSO or sequential threshold least squares, the algorithm filters out key features describing the system's dynamics from the feature library. Then, these features are used to build a dynamic model, typically a set of differential equations, describing how the physical field data changes with time and space. The model is then validated in the third step, and analysis is used to explore the physical mechanisms driving the changes in physical field information. The core advantage of this algorithm lies in its ability to reveal simplified dynamic laws from complex time-slice data, demonstrating powerful analytical capabilities even in scenarios where traditional modeling methods are difficult to apply.

[0061] 3. Nonlinear Modal Feature Recognition Based on Physical Information-Driven Neural Networks

[0062] Current fire simulation models and software require extensive experimental data to validate their solutions in practical applications, but these models often struggle to integrate high-fidelity data during the validation process. This is particularly problematic when solving inverse problems, where boundary conditions and fluid parameters are unknown. Model parameters must be calculated and the flow field reconstructed using measurement data from a limited number of locations, and current mainstream CFD methods suffer from poor accuracy. Furthermore, the finite element method demands high-quality mesh generation, and adjusting the mesh structure or even the mesh generation itself is extremely time-consuming.

[0063] In the context of multimodal problems, this method exhibits significant advantages, capable of handling complex phenomena in physics, engineering, and biology, typically described by a set of nonlinear partial differential equations involving multiple interacting physical processes or multiple temporal and spatial scales. Physically driven neural networks integrate deep neural networks with physical laws, providing a general solution strategy for complex nonlinear partial differential equation problems without relying on prior assumptions or linear simplifications. This method overcomes the limitations of traditional computational methods, solving not only problems dependent solely on initial and boundary conditions but also complex problems with internal structures by incorporating underlying physical laws. It can both acquire data distribution patterns through training samples, like traditional neural networks, and learn the physical laws behind the data. Compared to purely data-driven approaches, physics-driven neural networks, by using physical information to constrain the solution process during training, can more accurately learn the overall model's patterns.

[0064] The implementation of algorithms for physically-driven neural networks typically includes the following stages, such as... Figure 4 As shown: First, one or more neural networks are defined to approximate the solution of one or more variables of a partial differential equation. Then, during optimization, in addition to minimizing the difference between the network output and the actual observations, the residuals of the equations are further minimized by introducing residual terms derived from the physical equations. This process essentially ensures that the network output satisfies the considered physical laws. Specifically, this residual term is generated based on physical laws (e.g., the momentum equation, the mass conservation equation, etc.) and derivative information obtained through automatic differentiation techniques.

[0065] In real-world disaster scenarios, the development processes of various factors can often be described by multiple partial differential equations, whose analytical or numerical solutions contain a wealth of physical field information. Therefore, leveraging the efficient fitting capabilities of neural networks for nonlinear functions, a general function solver for partial differential equations can be constructed, transforming complex multimodal problems into problems involving nonlinear partial differential equations without requiring prior assumptions or linear simplifications to the model. By utilizing automatic differentiation and combining information such as the coordinates and parameters of the physical model to constrain the neural network, a neural network imbued with physical information can be obtained.

[0066] Meanwhile, as discussed in the previous step, the simulation of the integrated utility tunnel allows us to obtain the changes in physical information such as temperature, carbon monoxide, carbon dioxide, and oxygen concentrations under various conditions during a fire. This physical information is then used to construct a dataset (x, t, θ), where x represents the location of a point in space, t represents the time information at that moment, and θ represents the physical information of that point. For a given set of (x, t), the neural network will provide the corresponding θ solution. The nonlinear modal feature recognition process based on a physical information-driven neural network is as follows: Figure 5 As shown.

[0067] A trained, physically-driven neural network model is used to reconstruct the temperature field and perform modal analysis within the integrated utility tunnel area. On one hand, the model can reconstruct the spatial temperature field at any given time and deduce the temperature values ​​at the coordinates of each sensor location at that time. On the other hand, the model can identify the fire source development mode and temperature diffusion model parameters based on the temperature field and fire source intensity over a previous period, thereby obtaining the fire development status within the utility tunnel space for the next time period and enabling fire prediction.

[0068] 4. Rapid fire location based on Bayesian inference algorithm

[0069] In various current scenarios, solving inverse problems is becoming increasingly important. Essentially, solving inverse problems involves estimating the unknown parameters in a system of partial differential equations (PDEs) using a set of observations. In fire scenarios, the conditions for inverse problems are often incomplete or missing, as the solution may not exist or may not be unique. Furthermore, due to the relationship between parameters and observations, small perturbations in the data can lead to significant changes in the solution. However, in practice, observations are always noisy and their quantity or resolution may be limited. Therefore, quantifying the uncertainties introduced in the solution is often a challenge in inverse problem solving.

[0070] The Bayesian method is chosen to solve these problems. The Bayesian algorithm is essentially a method that transforms prior information into posterior information by continuously updating information, and it performs probabilistic deduction based on Bayes' theorem.

[0071]

[0072] Where x represents the model parameters, y represents the measured values, p(x|y) is the posterior probability density function of x, p(x) is the prior probability density function of x, which is generally derived from expert experience and subjective judgment, and p(y|x) is the likelihood function, representing the degree of fit between the model parameters and the observed data. The specific process is as follows... Figure 6 As shown.

[0073] (1) Prior probability

[0074] Prior information typically comes from experience or subjective judgment. For example, in fire identification within a utility tunnel, the location of the fire source is usually determined by directly selecting the length range of the tunnel. Prior information is often quite coarse and must be expressed as a probability density function before it can be used.

[0075] (2) Likelihood function

[0076] The likelihood function represents the degree of model parameters and is typically constructed as the normal distribution density function, exponential distribution density function, and Dirac function, etc., with the following expressions:

[0077] normal distribution:

[0078]

[0079] Laplace distribution:

[0080]

[0081] Dirac function:

[0082] p(y|x)=δ(yf(x))

[0083] Where, σ 2 Let be the variance of the noise, and n be the number of measurement data.

[0084] Since observation errors in fires are generally white noise, we assume the observation noise ε i Follows a normal distribution N(0,σ) 2 The likelihood function can be written as:

[0085]

[0086] According to the above formula, we can obtain:

[0087]

[0088] Where λ is a proportionality constant.

[0089] (3) Posterior probability density function sampling

[0090] Due to the complex mechanism of fire occurrence, it is difficult to deduce the system model. The results of the posterior distribution generally have a high dimension and poor analysis effect. To solve the high-dimensional inversion problem, the Metropolis-Hastings (M-H) sampling method is selected here to achieve sampling of the posterior probability density function. The M-H algorithm is the basis of the MCMC method, relying on the Markov process to achieve the purpose of sampling from the probability distribution function, and it is the most widely used MCMC sampling method at present.

[0091] To establish a Markov chain with π(x) as the stationary distribution, an arbitrary non-transfer probability q(x) and a function a(x), 0 < a(x) < 1, are selected. For any combination (x, x * ), a transition kernel is formed: p(x, x * ) = q(x, x * )α(x, x * ), x ≠ x * . If the Markov chain is at position x at time t, that is, X(t) = x, then q(x) will first generate a potential transition state x → x * , and then judge whether to transfer according to the acceptance probability α(x, x * ). That is, after finding the potential transfer point x * , if u ≤ α(x, x * ), then accept α(x, x * ) as the state value of the Markov chain at the next moment; if u > α(x, x * ), reject the candidate sample and maintain the current state. Thus, after having the transfer point x * , a number can be randomly and uniformly generated from the interval [0, 1], then:

[0092]

[0093] q(x) is generally called the proposal distribution. To make the posterior distribution π(x) stationary, α needs to be selected so that the corresponding p(x|x * ) has π(x) as the equilibrium distribution:

[0094]

[0095] At this time,

[0096]

[0097] When the expected distribution satisfies q(x, x * ) = q(x * , x), α(x, x * ) can be simplified to:

[0098]

[0099] The specific sampling steps are generally as follows:

[0100] 1. Initialization: t = 0, generate a random parameter x0 for the current x. t Assign a value and set the iteration termination step t = T.

[0101] 2. Let t = t + 1, then the conditional probability distribution θ(x|x t-1 Generate candidate samples x ′ .

[0102] 3. Calculate the acceptance probability α.

[0103] 4. Generate random numbers α from a uniform distribution. t .

[0104] 5. If α t If x ≤ α, then accept the candidate sample, and in this case, x t =x ′ Otherwise, reject the sample and let x t =x t-1 .

[0105] 6. Stop iterating when t = T, otherwise skip to step 2 and continue iterating.

[0106] (4) Positioning Algorithm

[0107] Based on the above, by using the fire development and physical information diffusion model extracted in the second step, as well as the corresponding characteristic parameters, the location of the fire ignition point can be identified by the maximum values ​​of parameters such as temperature, carbon dioxide, and carbon monoxide.

[0108] Example

[0109] 1) CFD numerical model of disaster scenarios in integrated utility tunnels

[0110] Based on the fire protection zoning of the integrated utility tunnel and the requirements of this invention, a 1:1 three-dimensional model was established according to the dimensions of the integrated utility tunnel. The model is 200m long, 3m wide, and 3m high. The ambient temperature was set to a fixed value of 20℃, and the air pressure was set to standard atmospheric pressure. Taking the fire source location as 0, the fire source level as 2kW, and the ventilation condition as 1m / s as an example, the numerical simulation results are as follows: Figure 7 As shown in (a) and (b).

[0111] 2) Physical mode reduction of time-slice data and physical information-driven nonlinear feature mode recognition

[0112] In the above model, slices are set up, and time slice data of physical field information such as location, temperature, carbon dioxide concentration, carbon monoxide concentration, and oxygen concentration are extracted from the model every 0.1 seconds. A point is randomly selected as a sampling point at other locations in the integrated utility tunnel except for the boundary area, and its location coordinates are determined. The sampling point location and related fire parameter information are used as the first type of initial training data. The random sampling process is repeated multiple times to obtain multiple sets of training data. Finally, a large amount of training data constitutes the first type of initial training dataset, including location information. The selection of the second type of training data is similar to the first type, except that the sampling location is converted to the boundary area. After repeated sampling, the second type of initial training dataset can be obtained.

[0113] Based on the relevant heat conduction equations and boundary conditions satisfied within the integrated utility tunnel, a first-class loss function driven by physics is constructed.

[0114] The differential equation for heat conduction is:

[0115]

[0116] The first type of loss function is:

[0117]

[0118] In the formula, w pde w bc For hyperparameters, L pde For the loss of the internal partial differential equation, L bc For boundary condition loss, P pde For the first type of initial training set, T θ The output temperature field is θ, where θ is the model parameter and φ is the model parameter. rated Let P be the heat source intensity distribution function. bc This is the second type of training set, where T0 is the temperature at the vent location or the ambient temperature, and n is the direction of the boundary normal.

[0119] Based on the heat conduction equations and boundary conditions within the integrated utility tunnel space, a second type of loss function, driven by physics and constrained by sensor data, is constructed. During training, the fire source intensity and spatial temperature field are solved as unknown parameters. Initial values ​​are obtained using the preliminary assessment of fire source intensity after data fusion. The loss function is minimized using selected training data to optimize the model, thereby fitting the correspondence between the measured temperature values ​​at the corresponding sensor locations and the temperature field of the utility tunnel space. Ultimately, a model conforming to the physical laws governing the development of fires in utility tunnels is obtained.

[0120] Based on the differential equation of heat conduction within the pipe gallery and the boundary conditions, the second loss function, driven by physics and constrained by sensor measurements, is set as follows:

[0121]

[0122] In the formula, w data For hyperparameters, L data For the loss term based on the temperature sensor monitoring value, T obs P is the temperature sensor reading. data This is the third type of training set.

[0123] Based on monitoring data from temperature sensors, a trained model is used to reconstruct the temperature field and perform modal analysis within the integrated utility tunnel area. On one hand, the model can reconstruct the spatial temperature field at any given time and deduce the temperature values ​​at the coordinates of each sensor location at that time. Simultaneously, the model can identify the fire source development mode and temperature diffusion model parameters based on the temperature field and fire source intensity over a previous period, thereby obtaining the fire development status within the utility tunnel space for the next time period and enabling fire prediction.

[0124] 3) Rapid fire location based on Bayesian inference algorithm

[0125] Based on temperature, the fire temperature characteristic parameters identified by the Bayesian fire estimation model were analyzed, including the average value and standard deviation during the sampling process, as shown in Table 1.

[0126] Table 1. Results of rapid fire location identification based on Bayesian inference algorithm

[0127]

[0128] The algorithm results show that the improved Bayesian inference algorithm can accurately estimate fire state parameters based on limited temperature monitoring values. Since the temperature above the fire source rises the most in a fire scenario compared to other areas, the fire source location can be identified relatively easily. When the fire source is located between two sensors, its location can also be identified based on the current airflow and the correlation between different sensors. The fire source location identified by the algorithm is almost identical to the actual location of the simulated fire source. Furthermore, the temperature in the pipe gallery remains relatively stable after 180 seconds, with minimal temperature increase, which is consistent with the temperature distribution in the fitted data, demonstrating the accuracy of the improved Bayesian inference algorithm in fire identification.

[0129] Taking sampling data from five typical working conditions as examples, the accuracy of tracing the location of the fire source is analyzed, such as... Figure 8 As shown, the improved Bayesian inference algorithm has higher accuracy and inference efficiency, and all sampling processes converge within 100 generations after the improvement. The fast fire location algorithm based on Bayesian inference can accurately trace the source of fires in integrated utility tunnels.

[0130] The spatial temperature distribution within the integrated utility tunnel can also be calculated using fire characteristic parameters estimated by an improved Bayesian inference algorithm. For example... Figure 9 As shown, when a fire occurs, the fire source is concentrated near the origin. As the fire develops, it spreads, the ambient temperature rises, and the fire's impact area gradually expands. When fires occur at different locations within the utility tunnel, the improved Bayesian algorithm can quickly locate the fire and, based on factors such as temperature field and ambient wind, achieve fire source tracing in the utility tunnel scenario. The algorithm's deduction results are similar to the numerical simulation results, and the temperature fields are essentially the same.

[0131] Based on the same inventive concept, this application provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the aforementioned integrated utility tunnel fire location method based on modal recognition and Bayesian algorithm.

[0132] Based on the same inventive concept, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the aforementioned integrated utility tunnel fire location method based on modal recognition and Bayesian algorithm.

[0133] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0134] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0135] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0136] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0137] The above embodiments are merely illustrative of the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made to the technical solutions based on the technical concept proposed in this invention shall fall within the scope of protection of this invention.

Claims

1. A method for locating fires in integrated utility tunnels based on modal recognition and Bayesian algorithms, characterized in that, Includes the following steps: Step 1: Construct a CFD numerical model of a fire scenario in a utility tunnel and solve the model to obtain the physical field information of the utility tunnel as it changes over time. Step 2: Based on the physical field information of the integrated utility tunnel that changes over time obtained in Step 1, construct time slice data, and use a nonlinear sparse identification algorithm to extract the physical field mode reduction model from the time slice data. Step 3: Construct a physical information neural network model that takes the location of the fire source and the corresponding temperature information as input and the predicted temperature field information inside the integrated utility tunnel as output. Obtain the first type of initial training set, the second type of training set and the third type of training set according to the physical field mode reduction model. Use the first type of initial training set to pre-train the physical information neural network model to obtain the pre-trained model. Further train the pre-trained model using the second and third type of training sets to obtain the trained model. Step 4: Reconstruct the temperature field within the integrated utility tunnel using the trained model, and quickly locate the fire source based on the reconstructed temperature field and Bayesian inference algorithm.

2. The integrated utility tunnel fire location method based on modal recognition and Bayesian algorithm according to claim 1, characterized in that, The specific process of step 1 is as follows: Based on different fire source locations, fire intensity, and ventilation conditions inside the utility tunnel, a CFD numerical model of a fire scenario in the utility tunnel is constructed using computational fluid dynamics (CFD) technology. The model is then solved using a semi-implicit algorithm based on the pressure coupling equations to obtain the physical field information of the utility tunnel that changes over time, including data on temperature, carbon monoxide concentration, carbon dioxide concentration, and oxygen concentration during a fire.

3. The integrated utility tunnel fire location method based on modal recognition and Bayesian algorithm according to claim 1, characterized in that, In step 2, based on the physical field information of the integrated utility tunnel that changes over time obtained in step 1, the slicing time is set to 0.1s, and time slice data of fire source location, temperature, carbon monoxide concentration, carbon dioxide concentration and oxygen concentration in the physical field information are extracted every 0.1s. The formula for the physical field modal order reduction model is as follows: In the formula, X is the state variable, Θ(X) is the feature library function matrix, and Θ1(X), Θ2(X), Θ... n (X) are all feature library functions, and Ξ is a sparse regression coefficient matrix used to describe the physical field modal reduction model, ∈1(X), ∈2(X), ∈ n (X) are all sparse regression coefficients.

4. The integrated utility tunnel fire location method based on modal recognition and Bayesian algorithm according to claim 3, characterized in that, The specific process of step 3 is as follows: A physical information neural network model is constructed, which takes the location of the fire source and the corresponding temperature information as input and the predicted internal temperature field information of the integrated utility tunnel as output. A point is randomly selected from other locations in the integrated utility tunnel except for the boundary area as the first sampling point. The location of the first sampling point and the corresponding physical field information are obtained as a training data in the first type of initial training set. The process of randomly selecting the first sampling point and obtaining the corresponding training data is repeated multiple times to generate the first type of initial training set. A point is randomly selected in the boundary area of ​​the integrated utility tunnel as the second sampling point. The location of the second sampling point and the corresponding physical field information are obtained as a training data in the second type of training set. The process of randomly selecting the second sampling point and obtaining the corresponding training data is repeated multiple times to generate the second type of training set. Based on the heat conduction equation and the boundary of the integrated utility tunnel, a first type of loss function is constructed. The physical information neural network model is pre-trained using the first type of initial training set and the first type of loss function to obtain the model parameters that minimize the first type of loss function, which are then used as the parameters of the pre-trained model. The heat conduction equation is as follows: In the formula, k is the thermal conductivity coefficient, T is the temperature, φ(x,y) is the heat source term, which represents the internal heat source or heat generated per unit volume, and (x,y) is the position coordinate; The formula for the first type of loss function is as follows: In the formula, Let w represent the first type of loss function. pde w bc For hyperparameters, L pde For the loss of the internal partial differential equation, L bc For boundary condition loss, P pde For the first type of initial training set, T θ The output temperature field is θ, where θ is the model parameter and φ is the model parameter. rated Let P be the heat source intensity distribution function. bc For the second type of training set, T0 is the temperature at the vent location or the ambient temperature, and n is the direction of the boundary normal. A second type of loss function is constructed based on the heat conduction equation and the boundary of the integrated utility tunnel. The pre-trained model is then trained again using the second type of training set and the second type of loss function to obtain the model parameters that minimize the second type of loss function, which are then used as the parameters of the trained model. The formula for the second type of loss function is as follows: In the formula, Let w represent the second type of loss function. data For hyperparameters, L data For the loss term based on the temperature sensor monitoring value, T obs P is the temperature sensor reading. data This is the third type of training set; the third type of training set consists of training data composed of the temperature monitored by temperature sensors deployed in the integrated utility tunnel and the location coordinates of the temperature sensors.

5. The integrated utility tunnel fire location method based on modal recognition and Bayesian algorithm according to claim 1, characterized in that, The specific process of step 4 is as follows: Step 4.1: Use the temperature, corresponding time, and location coordinates of the temperature sensors installed in the integrated utility tunnel as inputs to the Bayesian inference algorithm. Step 4.2: Assuming the fire source is located at the exact center of the utility tunnel, reconstruct the temperature field within the utility tunnel based on the model trained in Step 3. Step 4.3: Calculate the likelihood probability of the temperature monitored by each temperature sensor under the temperature field reconstructed in step 4.2; Step 4.4: Based on the likelihood probability and the temperature monitored by the temperature sensor, the location of the fire source is calculated using the Markov chain Monte Carlo method. Step 4.5: Determine whether the fire source location calculated in Step 4.4 is consistent with the fire source location set in Step 4.

2. If so, use the fire source location calculated in Step 4.4 as the final fire source location output; otherwise, proceed to Step 4.

6. Step 4.6: Replace the fire source location set in step 4.2 with the fire source location calculated in step 4.4, and repeat steps 4.2 to 4.

5.

6. A computer device comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the integrated utility tunnel fire location method based on modal recognition and Bayesian algorithm as described in any one of claims 1 to 5.

7. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the integrated utility tunnel fire location method based on modal recognition and Bayesian algorithm as described in any one of claims 1 to 5.

Images