Fire prediction method and system based on data assimilation coupled physical information neural network

By using data assimilation in the tunnel and combining with limited sensor data, fast, accurate and real-time prediction of the entire flow field of the tunnel fire smoke is achieved, and the insufficient monitoring coverage and high computational complexity are solved in traditional methods. The problem has improved the efficiency of fire monitoring and emergency response.

CN118395368BActive Publication Date: 2024-11-15CHONGQING UNIV
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
CN202410327326.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-03-21
Publication Date
2024-11-15
Estimated Expiration
2044-03-21

AI Technical Summary

Technical Problem

Existing tunnel fire monitoring technology cannot accurately predict the fire development process, the sensor coverage is insufficient, the CFD numerical simulation calculation is complex and time-consuming, making it difficult to achieve real-time monitoring and forecasting, and the prediction of the heat release rate of the fire source is inaccurate.

Method used

Using a method of coupled physical information neural network based on data assimilation, the fire source is located through the limited sensor data in the tunnel, data assimilation and neural network prediction are achieved to achieve fast, accurate and real-time prediction of the entire flow field of the tunnel fire smoke.

Benefits of technology

Through a small number of fixed temperature sensors, the rapid, accurate and real-time prediction of the entire flow field of the tunnel fire flue gas is achieved, which improves the fire monitoring, early warning and emergency response capabilities, reduces the cost of system construction and commissioning, and improves the computing efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118395368B_ABST
    Figure CN118395368B_ABST
Patent Text Reader

Abstract

The present invention discloses a fire prediction method based on data assimilation coupled physical information neural network, including: S1, when a fire occurs in a tunnel, locate the fire source; S2, arrange temperature sensors in different layouts in the tunnel, obtain the system state parameters of the fire smoke by data assimilation of the time series temperature data collected by the temperature sensors in different layouts, calculate the temperature prediction value of the tunnel fire smoke, and obtain the temperature prediction accuracy. When the temperature prediction accuracy is less than or equal to the set threshold, this layout is the optimal layout of the temperature sensor; S3, based on the time series temperature data and system state parameters collected by the temperature sensor in the optimal layout, predict the temperature field, velocity field and pressure field of the tunnel fire smoke based on data assimilation coupled physical information neural network. The present invention realizes fast, accurate and real-time prediction of the flow field of the whole tunnel fire smoke field through a small amount of fixed temperature sensor data.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to a fire prediction method, in particular to a fire prediction method based on data assimilation coupled physical information neural network, and also to a system using the fire prediction method. Background Art

[0002] Tunnel fire is a very destructive accident that often results in casualties and losses. Currently, tunnel fire monitoring technology mainly relies on traditional sensor layout and finite element analysis methods. However, traditional methods are limited by parameter uncertainty and measurement data limitations and cannot accurately predict the development of fire.

[0003] Traditional tunnel fire monitoring and prediction methods have the following main defects: the monitoring coverage of sensors is insufficient, and they can only obtain temperature information at local locations, but cannot predict the temperature field, velocity field and pressure field of the entire field; CFD numerical simulation has high computational complexity and relies on time and computer resources, making it difficult to monitor and predict the development of fire in real time; there is a lack of accurate prediction of key parameters such as the heat release rate of the fire source.

[0004] At present, some methods try to achieve real-time prediction of smoke flow field by combining numerical simulation with artificial neural network. However, this method requires a large amount of numerical simulation data to train the neural network, which has certain requirements on computer resources and time. Moreover, for different types of tunnels, the data set needs to be rebuilt and adjusted, which increases the cost and complexity of system construction and debugging. Summary of the invention

[0005] The first object of the present invention is to provide a fire prediction method based on data assimilation coupled physical information neural network, which can realize fast, accurate and real-time prediction of the flow field of the whole tunnel fire smoke through limited sensors in the tunnel.

[0006] The first object of the present invention is achieved by the following technical measures: A fire prediction method based on data assimilation coupled physical information neural network, characterized by comprising the following steps:

[0007] S1. When a fire occurs in a tunnel, locate the fire source;

[0008] S2. Arrange temperature sensors in the tunnel in different arrangements, obtain the system state parameters of the fire smoke through data assimilation of the time series temperature data collected by the temperature sensors in different arrangements, calculate the temperature prediction value of the tunnel fire smoke, and obtain the temperature prediction accuracy. When the temperature prediction accuracy is less than or equal to the set threshold, this arrangement is the optimal arrangement of the temperature sensors;

[0009] S3. According to the time series temperature data and system state parameters collected by the temperature sensors under the optimal arrangement, the temperature field, velocity field and pressure field of the tunnel fire smoke are predicted based on data assimilation and coupled physical information neural network.

[0010] When a tunnel fire occurs, the present invention can realize rapid, accurate and real-time prediction of the flow field of the entire tunnel fire smoke through a small amount of fixed temperature sensor data, thereby improving the monitoring, early warning and emergency response capabilities of the tunnel fire.

[0011] The step S2 of the present invention comprises:

[0012] (1) Arrange the temperature sensors in a certain arrangement to obtain the time series temperature data {x, y, t, T} of the temperature sensors;

[0013] (2) Use the Monte-Carlo method to randomly generate the initial uncertainty coefficient set λ i f and the fire source heat release rate set q f , the probability model satisfied by all set members is normal distribution;

[0014] ⑶The formulas describing heat transfer, mass transfer and fluid dynamics processes in fire are as follows:

[0015]

[0016] Where: u is the velocity field;

[0017]

[0018] Where: ρ is density; p is pressure; μ is dynamic viscosity; f is external force;

[0019]

[0020] Where: T is temperature; α is the thermal diffusion coefficient, which indicates the thermal conductivity of the medium; q is the heat source;

[0021] p=ρRT / W Formula (4)

[0022] Where: R is the gas constant; W is the molar mass of the mixed gas;

[0023] Transforming formulas (1) to (4), we obtain:

[0024]

[0025]

[0026]

[0027] ⑷ Set the fire source position (x0, y0), sensor position (x, y, t), and uncertainty coefficient and the fire source heat release rate set q f Substitute into formula (5)-(7) and solve to obtain the temperature prediction value T predicton ;

[0028] ⑸ According to the temperature prediction value T predicton Calculate the optimized system state parameters with the time series temperature data {x,y,t,T} collected by the temperature sensor q f , and then the fire source position (x0, y0), sensor position (x, y, t) and optimized system state parameters q f Substitute into formula (5)-(7) and solve to obtain the temperature prediction value T predicton Repeat this step to perform data assimilation. When the number of assimilation times is greater than the set value, the temperature prediction value T predicton , calculate the temperature prediction accuracy, that is, the temperature prediction value T predicton The mean square error (MSE) between the time series temperature data is:

[0029]

[0030] Where: Y i is the actual observed value, that is, the time series temperature data; is the corresponding model prediction value;

[0031] When the MSE is greater than the set threshold value I1, the process proceeds to step (1) to arrange the temperature sensors in another arrangement mode;

[0032] When the MSE is less than or equal to the set threshold I1, the arrangement of the temperature sensor is the optimal arrangement, and the time series temperature data {x, y, t, T} of the temperature sensor under the optimal arrangement and the corresponding system state parameters λ1, λ2, q are obtained.

[0033] The threshold value I1 is set as [0.01-0.1] in the present invention.

[0034] Step S3 of the present invention comprises:

[0035] (1) Input the time series temperature data {x, y, t, T} of the sensor under the optimal arrangement as the labeled training data into the fully connected feedforward neural network for training, and obtain the weight w and bias b in the feedforward neural network;

[0036] (2) The automatic differentiation feature of deep learning is used to simplify the calculation of partial derivatives. The calculated partial differential values ​​and the system state parameters λ1, λ2, q under the optimal layout are substituted into the following formulas (9)-(11) to obtain the loss function L of the physical model part: PDE and the model's predicted value

[0037] The loss function L of the physical model part PDE It is defined as the residual of formula (5)-(7) on the labeled training data and is expressed as:

[0038]

[0039]

[0040]

[0041] Where: T is the output of the neural network during each iteration; f i (i=1,2,3) is a soft constraint on the learning process and is incorporated into the loss function;

[0042] Get the loss function of the physical model part:

[0043]

[0044] ⑶ Based on the labeled training data and the predicted value of the model Calculate the loss function L of the data-driven part Data :

[0045]

[0046] (4) Calculate the comprehensive loss function L:

[0047] L=w Data L data +w PDE L PDE Formula (14)

[0048] Where: w Data and w PDE are the corresponding weights of the data-driven part and the physical model part;

[0049] (5) When the comprehensive loss function L is greater than the set threshold value I2, the updated unknown parameters w, b, λ1, λ2, q are input into the feedforward neural network and formulas (9)-(11) to obtain the loss function of the physical model part, and then proceed to step (3);

[0050] When the comprehensive loss function L is less than or equal to the set threshold I2, the final parameters λ′1, λ′2, q′ and the space-time coordinates {x′, y′, t} are substituted into formulas (5)-(7) to obtain the temperature, pressure and velocity at each space-time coordinate, that is, the temperature field, velocity field and pressure field of the tunnel fire smoke.

[0051] The threshold value I2 is set as described in the present invention = (0-0.01].

[0052] In the step S2⑸ of the present invention, the number of data assimilation is 10-100.

[0053] The second object of the present invention is to provide a system using the above-mentioned fire prediction method based on data assimilation coupled physical information neural network.

[0054] The second object of the present invention is achieved through the following technical measures: a system using the above-mentioned fire prediction method based on data assimilation coupled physical information neural network, characterized in that it includes a temperature sensor, a data acquisition device, a server integrated with data assimilation and physical information neural network algorithms, a display device, a control device and a fire-fighting device installed in the tunnel, wherein the temperature sensor is arranged along the longitudinal center axis of the tunnel, and the temperature sensor transmits the collected time series temperature data to the data acquisition device, which is processed and sent to the server, and the server calculates the predicted results of the temperature field, velocity field and pressure field of the tunnel fire smoke and displays them on the display device, and the control device provides a fire-fighting strategy reference and controls the fire-fighting device according to the prediction results.

[0055] The temperature sensor of the present invention is no more than 20 cm away from the ceiling.

[0056] In the present invention, there are no more than two temperature sensors at the same longitudinal position of the tunnel.

[0057] The interval between adjacent temperature sensors of the present invention is 20-50m.

[0058] Compared with the prior art, the present invention has the following significant effects:

[0059] (1) When a tunnel fire occurs, the present invention can use a small amount of fixed temperature sensor data to achieve rapid, accurate and real-time prediction of the flow field of the entire tunnel fire smoke field, thereby improving the monitoring, early warning and emergency response capabilities of the tunnel fire.

[0060] ⑵The present invention combines sensor data and mathematical models, embeds physical equations as constraints into the neural network, makes full use of real-time observation data and physical laws, ensures that the prediction results conform to actual physical laws, and improves the accuracy and reliability of fire monitoring.

[0061] ⑶ The present invention has higher computing efficiency and parallelism, can complete the prediction task of complex systems in a shorter time, and improves the efficiency of monitoring and emergency response. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0063] Figure 1 It is a prediction flow chart of the method of the present invention;

[0064] Figure 2 It is a flow chart of the optimal system state parameter estimation based on data assimilation of the present invention;

[0065] Figure 3 It is a flow chart of tunnel fire process prediction based on physical information network of the present invention;

[0066] Figure 4 It is a schematic diagram of the composition of the system of the present invention. DETAILED DESCRIPTION

[0067] The present invention is described in detail below in conjunction with the embodiments and the accompanying drawings to help those skilled in the art better understand the inventive concept of the present invention. However, the protection scope of the claims of the present invention is not limited to the following embodiments. For those skilled in the art, all other embodiments obtained without creative work without departing from the inventive concept of the present invention belong to the protection scope of the present invention.

[0068] like Figures 1 to 3 As shown, the present invention provides a fire prediction method based on data assimilation coupled physical information neural network, comprising the following steps:

[0069] S1. When a fire occurs in the tunnel, the fire source is located (x0, y0). Specifically, when an abnormal temperature rise is detected in the tunnel, the fire source is located through sensors and video monitoring.

[0070] S2. Arrange temperature sensors in the tunnel in different arrangements, obtain the system state parameters of the fire smoke through data assimilation of the time series temperature data collected by the temperature sensors in different arrangements, calculate the temperature prediction value of the tunnel fire smoke, and obtain the temperature prediction accuracy. When the temperature prediction accuracy is less than or equal to the set threshold, this arrangement is the optimal arrangement of the temperature sensors;

[0071] Specifically include:

[0072] (1) Arrange the temperature sensors in a certain arrangement to obtain the time series temperature data {x, y, t, T} of the temperature sensors;

[0073] (2) Use the Monte-Carlo method to randomly generate the initial uncertainty coefficient set and the fire source heat release rate set q f , the probability model satisfied by all set members is normal distribution;

[0074] ⑶ When a fire occurs in a tunnel, the following assumptions are made: (1) two-dimensional flow; (2) smoke is an incompressible fluid (i.e., the volume of the microelement is assumed to be constant); (3) when the volume is constant, ρ does not change with T. (4) The evolution of the smoke flow temperature field is mainly affected by heat conduction, heat convection and heat source. The heat source in a tunnel fire mainly considers the fire source position and the heat release rate per unit area. When the fire source position (x0, y0) is known, the fire heat release rate can be regarded as a source term q that varies with time.

[0075] The heat transfer, mass transfer and fluid dynamics processes in fire are described using partial differential equations:

[0076] Continuity equation (mass equation):

[0077]

[0078] Where: u is the velocity field;

[0079] Momentum equation:

[0080]

[0081] Where: ρ is density; p is pressure; μ is dynamic viscosity; f is external force;

[0082] Energy equation:

[0083]

[0084] Where: T is temperature; α is the thermal diffusion coefficient, which indicates the thermal conductivity of the medium; q is the heat source;

[0085] Ideal gas state equation:

[0086] p=ρRT / W Formula (4)

[0087] Where: R is the gas constant; W is the molar mass of the mixed gas;

[0088] When a fire occurs in a tunnel, the boundary conditions such as the heat release rate of the fire source are unknown, which leads to the fact that q in the energy equation formula (4) is actually unknown. In addition, the physical parameters μ and α of air change with temperature. Therefore, by transforming formulas (1) to (4), we get:

[0089]

[0090]

[0091]

[0092] ⑷ Set the fire source position (x0, y0), sensor position (x, y, t), and uncertainty coefficient set λ i f and the fire source heat release rate set q f Substitute into formula (5)-(7) and solve to obtain the temperature prediction value T predicton ;

[0093] ⑸ The predicted temperature value T predicton The time series temperature data {x, y, t, T} collected by the temperature sensor is used to obtain the optimized system state parameters through the Kalman gain in data assimilation. q f , and then the fire source position (x0, y0), sensor position (x, y, t) and optimized system state parameters q f Substitute into formula (5)-(7) and solve to obtain the temperature prediction value T predicton Repeat this step to perform data assimilation. When the number of assimilation times is greater than the set value, the temperature prediction value T predicton , calculate the temperature prediction accuracy, that is, the temperature prediction value T predicton The mean square error (MSE) between the time series temperature data is:

[0094]

[0095] Where: Y i is the actual observed value, that is, the time series temperature data; is the corresponding model prediction value;

[0096] Among them, data assimilation is mainly carried out through the EnKF method, which is an inverse modeling method that can correct the model state or parameters when observations are available.

[0097] X a =X f +K x,y (y-HX f )

[0098] Where: X a is composed of m analysis value vectors The analysis value set matrix composed of f is composed of m forecast value vectors The forecast ensemble matrix composed of ; y is the measured value; H is the observation operator, which contains the relative position relationship between the measured value and the predicted value vector; K x,yis the Kalman gain matrix, which is used to modify the model state and parameters.

[0099] In EnKF, K x,y It is statistically determined by multiple set members, similar to the Monte Carlo method, and is obtained by the following formula:

[0100]

[0101] Where: P is the prediction error covariance; R is the observation error, which is set by the user and is set to 5% of the observation in the following experiment; P is calculated by the following formula:

[0102]

[0103]

[0104] Where: x f is the average value of the forecast ensemble matrix. After obtaining the best estimated model state and model parameters, the forecast simulation can be continued and the subsequent time steps can be predicted.

[0105] When the MSE is greater than the set threshold value I1, the threshold value I1 is set to [0.01-0.1], and the process goes to step (1), and the temperature sensors are arranged in another arrangement mode;

[0106] When the MSE is less than or equal to the set threshold I1, the arrangement of the temperature sensor is the optimal arrangement, and the time series temperature data {x, y, t, T} of the temperature sensor under the optimal arrangement and the corresponding system state parameters λ1, λ2, q are obtained.

[0107] S3. According to the time series temperature data and system state parameters collected by the temperature sensors under the optimal arrangement, the temperature field, velocity field and pressure field of the tunnel fire smoke are predicted based on data assimilation and coupled physical information neural network.

[0108] Specifically include:

[0109] ⑴ Input the time series temperature data {x, y, t, T} of the temperature sensor under the optimal arrangement as the labeled training data into the fully connected feedforward neural network for training, and obtain the weight w and bias b in the feedforward neural network; the fully connected feedforward neural network generally includes an input layer, an output layer and n hidden layers, and the connection between the layers is expressed as follows:

[0110] y i =σ(w i ·y i-1 +b i )1≤i≤n

[0111] z=w n+1·y n +b n+1

[0112] Where: i = [1, n] represents the hidden layer, the n+1th layer is the output layer; y i represents the output of the i-th layer; z is the output of the neural network, which is the temperature T in our model. σ(·) represents the activation function, which allows the neural network to map nonlinear relationships. (w i ,b i ) represents the weights and biases of the i-th layer, which are updated during training.

[0113] (2) The automatic differentiation feature of deep learning is used to simplify the calculation of partial derivatives. The calculated partial differential values ​​and the system state parameters λ1, λ2, q under the optimal layout are substituted into the following formulas (9)-(11) to obtain the loss function L of the physical model part: PDE and the model's predicted value

[0114] The loss function L of the physical model part PDE It is defined as the residual of formula (5)-(7) on the labeled training data and is expressed as:

[0115]

[0116]

[0117]

[0118] Where: T is the output of the neural network during each iteration; f i (i=1,2,3) is a soft constraint on the learning process and is incorporated into the loss function;

[0119] Get the loss function of the physical model part:

[0120]

[0121] ⑶ Based on the labeled training data and the predicted value of the model Calculate the loss function L of the data-driven part Data :

[0122]

[0123] (4) Calculate the comprehensive loss function L:

[0124] L=w Data L data +w PDE L PDE Formula (14)

[0125] Where: wData and w PDE are the corresponding weights of the data-driven part and the physical model part;

[0126] (5) When the comprehensive loss function L is greater than the set threshold value I2, the threshold value I2 is set to (0-0.01], and the updated unknown parameters w, b, λ1, λ2, q are input into the feedforward neural network and formulas (9)-(11) to obtain the loss function of the physical model part, and then proceed to step (3);

[0127] When the comprehensive loss function L is less than or equal to the set threshold I2, the final parameters λ′1, λ′2, q′ and the space-time coordinates {x′, y′, t} are substituted into formulas (5)-(7) to obtain the temperature, pressure and velocity at each space-time coordinate, that is, the temperature field, velocity field and pressure field of the tunnel fire smoke.

[0128] See also Figure 4 , a system using the above-mentioned fire prediction method based on data assimilation coupled physical information neural network, including a temperature sensor (sparse temperature sensor), a data acquisition device (sensor data acquisition device), a server integrated with data assimilation and physical information neural network algorithms, a display device (fire smoke flow field prediction and display system), a control device and a fire-fighting device installed in the tunnel, the temperature sensor is arranged along the longitudinal center axis of the tunnel, the temperature sensor is no more than 20cm away from the ceiling, and the adjacent temperature sensors are spaced 20-50m apart. There are no more than 2 temperature sensors at the same longitudinal position in the tunnel. The temperature sensor transmits the collected time series temperature data to the data acquisition device, which is processed and sent to the server, and the temperature field, velocity field and pressure field of the tunnel fire smoke are predicted by the server, and the prediction results are displayed on the display device. The control device provides a fire-fighting strategy reference and controls the fire-fighting device according to the prediction results.

Claims

1. A fire prediction method based on data assimilation coupled physical information neural network, characterized in that The following steps are involved: S1. When a fire occurs in a tunnel, locate the fire source; S2. Arrange temperature sensors in the tunnel in different arrangements, obtain the system state parameters of the fire smoke through data assimilation of the time series temperature data collected by the temperature sensors in different arrangements, calculate the temperature prediction value of the tunnel fire smoke, and obtain the temperature prediction accuracy. When the temperature prediction accuracy is less than or equal to the set threshold, this arrangement is the optimal arrangement of the temperature sensors; (1) Arrange the temperature sensors in a certain arrangement to obtain the time series temperature data {x, y, t, T} of the temperature sensors; (2) Use the Monte-Carlo method to randomly generate the initial uncertainty coefficient set λ i f and the fire source heat release rate set q f , the probability model satisfied by all set members is normal distribution; ⑶The formulas describing heat transfer, mass transfer and fluid dynamics processes in fire are as follows: Where: u is the velocity field; Where: ρ is density; p is pressure; μ is dynamic viscosity; f is external force; Where: T is temperature; α is the thermal diffusion coefficient, which indicates the thermal conductivity of the medium; q is the heat source; p=ρRT / W Formula (4) Where: R is the gas constant; W is the molar mass of the mixed gas; Transforming formulas (1) to (4), we obtain: ⑷ Set the fire source position (x0, y0), sensor position (x, y, t), and uncertainty coefficient and the fire source heat release rate set q f Substitute into formula (5)-(7) and solve to obtain the temperature prediction value T predicton ; ⑸ According to the temperature prediction value T predicton Calculate the optimized system state parameters with the time series temperature data {x,y,t,T} collected by the temperature sensor q f , and then the fire source position (x0, y0), sensor position (x, y, t) and optimized system state parameters q f Substitute into formula (5)-(7) and solve to obtain the temperature prediction value T predicton Repeat this step to perform data assimilation. When the number of assimilation times is greater than the set value, the temperature prediction value T predicton , calculate the temperature prediction accuracy, that is, the temperature prediction value T predicton The mean square error (MSE) between the time series temperature data is: Where: Y i is the actual observed value, that is, the time series temperature data; is the corresponding model prediction value; When the MSE is greater than the set threshold value I1, the process proceeds to step (1) to arrange the temperature sensors in another arrangement mode; When MSE is less than or equal to the set threshold I1, the arrangement of the temperature sensor is the optimal arrangement, and the time series temperature data {x, y, t, T} of the temperature sensor under the optimal arrangement and the corresponding system state parameters λ1, λ2, q are obtained; S3, based on the time series temperature data and system state parameters collected by the temperature sensors under the optimal arrangement mode, the temperature field, velocity field and pressure field of the smoke from the tunnel fire are predicted based on the data assimilation coupled physical information neural network; (1) Input the time series temperature data {x, y, t, T} of the temperature sensor under the optimal arrangement as the labeled training data into the fully connected feedforward neural network for training, and obtain the weight w and bias b in the feedforward neural network; (2) The automatic differentiation feature of deep learning is used to simplify the calculation of partial derivatives. The calculated partial differential values ​​and the system state parameters λ1, λ2, q under the optimal layout are substituted into the following formulas (9)-(11) to obtain the loss function L of the physical model part: PDE and the model's predicted value The loss function L of the physical model part PDE It is defined as the residual of formula (5)-(7) on the labeled training data and is expressed as: Where: T is the output of the neural network during each iteration; f i (i=1,2,3) is a soft constraint on the learning process and is incorporated into the loss function; Get the loss function of the physical model part: ⑶ Based on the labeled training data and the predicted value of the model Calculate the loss function L of the data-driven part Data : (4) Calculate the comprehensive loss function L: L=w Data L data +w PDE L PDE Formula (14) Where: w Data and w PDE are the corresponding weights of the data-driven part and the physical model part; (5) When the comprehensive loss function L is greater than the set threshold value I2, the updated unknown parameters w, b, λ1, λ2, q are input into the feedforward neural network and formulas (9)-(11) to obtain the loss function of the physical model part, and then proceed to step (3); When the comprehensive loss function L is less than or equal to the set threshold I2, the final parameters λ′1, λ′2, q′ and the space-time coordinates {x′, y′, t} are substituted into formulas (5)-(7) to obtain the temperature, pressure and velocity at each space-time coordinate, that is, the temperature field, velocity field and pressure field of the tunnel fire smoke.

2. The fire prediction method based on data assimilation coupled physical information neural network according to claim 1 is characterized by: The set threshold I1 = [0.01-0.1].

3. The fire prediction method based on data assimilation coupled physical information neural network according to claim 2 is characterized by: The set threshold value I2=(0-0.01].

4. The fire prediction method based on data assimilation coupled physical information neural network according to claim 3 is characterized by: In step S2⑸, the number of data assimilation is 10-100.

5. A system using the fire prediction method based on data assimilation coupled physical information neural network according to any one of claims 1 to 4, characterized in that: It includes a temperature sensor, a data acquisition device, a server integrated with data assimilation and physical information neural network algorithms, a display device, a control device and a fire-fighting device installed in a tunnel. The temperature sensor is arranged along the longitudinal center axis of the tunnel. The temperature sensor transmits the collected time series temperature data to the data acquisition device, which processes and sends it to the server. The server predicts the temperature field, velocity field and pressure field of the smoke in the tunnel fire, and displays the prediction results on the display device. The control device provides a fire-fighting strategy reference and controls the fire-fighting device according to the prediction results.

6. The system according to claim 5, characterized in that: There shall be no more than 2 temperature sensors at the same longitudinal position in the tunnel.

7. The system according to claim 6, characterized in that: The temperature sensor is no more than 20 cm away from the ceiling.

8. The system according to claim 7, characterized in that: The interval between adjacent temperature sensors is 20-50m.

Citation Information

Patent Citations

  • Aircraft engine damage detection method

    CN109883708A

  • Numerical weather forecast mixed data assimilation method based on triple multi-layer perceptrons

    CN111783361A

  • Method and system for forecasting tunnel fire smoke ceiling jet flow development based on data assimilation

    CN115526123A

  • Space-time dynamic system soft measurement method for automatically determining partial differential equation structure

    CN116910428A

  • Integration of physical sensors in a data assimilation framework

    US20210199842A1