Simulation and data combined driving coal mine fire disaster dynamic disaster avoidance route planning method

Abstract

The invention provides a simulation and data combined driving coal mine fire disaster dynamic disaster avoidance route planning method, and belongs to the technical field of coal disaster prevention. The technical problems that lag is caused by traditional timing updating, multiple disaster avoidance paths cannot be planned for people in different areas at the same time, and the risk that people cannot pass through a single path is caused are solved. According to the technical scheme, S1, the position of a fire source is positioned; s2, distinguishing a safe area, an affected area and a high-danger area; s3, monitoring a roadway damage condition and updating the topological graph; s4, establishing a fire hazard and toxic and harmful substance spreading prediction model; s5, predicting the change trend dynamic of each factor influencing escape; s6, outputting optimal paths of different areas; and S7, sending the optimal path. The method has the beneficial effects that a plurality of disaster avoidance paths can be planned for personnel in different areas, the passing speed and weight of each roadway are updated in real time when a disaster occurs, the disaster avoidance paths are dynamically adjusted, and the escape success rate of miners is improved.

Description

Technical Field

[0001] This application relates to the technical field of coal mine disaster prevention, and particularly to a method for dynamically planning escape routes for coal mine fires jointly driven by simulation and data. Background Art

[0002] Mine fires are one of the five major disasters in coal mines. Fire accidents are characterized by a large impact range and severe disasters. Due to the particularity and complexity of the coal mine working environment, more serious accident consequences will occur when accidents occur in coal mines. Therefore, safety issues are of crucial importance in the coal mine industry. In related technologies, traditional escape methods are fixed routes formulated in advance based on the relevant experience of experts and staff, which cannot adapt to the dynamic changes of fires, thus affecting the safety of miners. Most of the existing escape route planning methods are static route planning, which also cannot effectively respond to the dynamic changes of fires. And the existing dynamic escape route planning updates the path at set intervals, which has a certain lag. Therefore, in order to increase the likelihood of survival of coal mine workers during fire accidents, a method for dynamically planning escape routes for coal mine fires jointly driven by simulation and data is proposed.

[0003] Patent application publication number CN117217400A discloses an escape route dynamic planning method, device, electronic device and storage medium. By generating a weighted undirected graph, combining the real-time position information of underground personnel and the escape location, the optimal escape route is calculated using the shortest path algorithm. This method can dynamically update the traffic influence factors to cope with the changes in disasters and adjust the escape route accordingly. In addition, the system can also push escape reminder information to underground personnel through voice or terminal devices to help them take timely evasive actions and improve the escape efficiency and safety of mine personnel. However, it has the following defects:

[0004] 1. The escape route is adjusted by updating the traffic influence factors at preset times, and this timed update is not sufficient to quickly respond to the dynamic changes of disasters;

[0005] 2. It ignores possible emergencies during the escape process such as roadway collapses, and the impact of cumulative lethal factors on escape;

[0006] 3. It mainly focuses on planning the shortest path from the current position to the escape location, but in the actual disaster environment, multiple alternative escape routes may be required. Summary of the Invention

[0007] Object of the Invention: Aiming at the above problems, the object of the present invention is to provide a method for dynamically planning escape routes for coal mine fires jointly driven by simulation and data to solve the problem in the prior art that the shortest escape route cannot be dynamically and real-time updated during fire accidents.

[0008] Technical solution: A method for dynamically planning an evacuation route for coal mine fires driven by simulation and data integration, comprising the following steps:

[0009] S1: Obtain data on gas concentration, temperature, and wind speed using fire sensors arranged in the roadway network. After data processing, use a spatio-temporal Transformer prediction model combined with a physical model inversion method to locate the fire source.

[0010] S2: Construct a directed weighted topological graph of the roadway according to the principle of graph theory. Mark the roadway where the fire source is located in the graph and set the weight to ∞, and distinguish the safe area, affected area, and high-risk area. The affected area includes a mild-risk area and a moderate-risk area.

[0011] S3: Monitor the roadway damage condition to update the topological graph. At the initial moment of the fire, use the k-optimal path method to plan k paths for the personnel in each area.

[0012] S4: After processing and fusing the parameter data samples obtained from fire simulation and the on-site parameter data samples, input them into the ConvLSTM prediction model to establish a prediction model for the spread of fire and toxic and harmful substances, so as to realize real-time dynamic evacuation route planning.

[0013] S5: Input the data monitored by the sensors during the fire into the prediction model to predict the dynamic change trends of the various factors affecting escape, and update the personnel escape speed and roadway weight.

[0014] S6: Set the constraints for the evacuation path, and use the ant colony algorithm to update the previous k paths, and output the optimal paths for different areas.

[0015] S7: Send the optimal path information to the mobile terminals carried by the workers and the underground signs.

[0016] Further, in step S1, the method of obtaining data on gas concentration, temperature, and wind speed using fire sensors arranged in the roadway network, and after data processing, using a spatio-temporal Transformer prediction model combined with a physical model inversion method to locate the fire source includes the following steps:

[0017] S11: Construct a spatio-temporal Transformer model, use historical fire data to generate training samples with the marked fire source location and intensity, define the loss function L Transformer = ||X true - X DL || + ||Q true - Q DL || and add the AdamW optimizer to establish a spatio-temporal Transformer model for preliminary fire source location.

[0018] S12. Perform Gaussian filtering denoising, linear time interpolation, spatial interpolation, and data normalization on the temperature T(x, y, z, t), gas concentration C(x, y, z, t), and wind speed V(x, y, z, t) data collected by the sensor to obtain the time series X = {T(x, y, z, t), C(x, y, z, t), V(x, y, z, t)}. Divide the data into multiple blocks and input them into the model. Use the self-attention mechanism to calculate the spatial features, capture the spatial relationship between sensors, and output the preliminary fire source location X DL and the fire source intensity Q DL ;

[0019] The Gaussian filtering formula is

[0020] The linear time interpolation formula is

[0021] The spatial interpolation formula is

[0022] The data normalization formula is

[0023] where

[0024] exp() is the exponential function with base e, T'(x, y) is the filtered data, G(i, j) is the two-dimensional Gaussian kernel, σ is the Gaussian filtering standard deviation, k is the filtering window radius, t1 and t2 are the two known time points on the left and right of the missing point, T(t1) and T(t2) are the corresponding known data values, T(t) is the missing point value, X(x, y, z) is the value of the interpolation point, X i the value of the known point i, d i is the distance between the interpolation point and the known point i, p is the power parameter, X is the original data, Xmin and Xmax are the maximum and minimum values of the data, and Xnorm is the normalized data;

[0025] S13. Substitute the preliminary fire source location X DL and the fire source intensity Q DL as the initial values of the physical model into the heat conduction model and the gas diffusion model to correct the deep learning prediction results;

[0026] The expression of the heat conduction model is

[0027] The expression of the gas diffusion model is

[0028] where

[0029] C is the gas concentration, D is the diffusion coefficient, V is the ventilation speed, S(x, y, z) is the gas source term, representing the gas release intensity at the fire source, T is the temperature, α is the diffusion coefficient, and Q(x, y, z) is the heat source term. is the Laplacian operator of the concentration, describing the diffusion behavior of the gas. is the convection term of the gas. is the Laplacian operator of the temperature;

[0030] S14, defining an error function based on sensor data and physical model results

[0031] Minimize the error function minE(X source , Q source ) by the gradient descent method to optimize the fire source location and intensity until |E (k+1) - E (k) | ≤ δ to terminate the iteration, and output the finally corrected X final and Q final ;

[0032] Among them,

[0033] X source is the fire source location, Q source is the fire source intensity, T obs,i is the sensor-observed temperature, T phys,i is the temperature calculated by the physical model, C obs,i is the sensor-observed gas concentration, C phys,i is the gas concentration calculated by the physical model.

[0034] The iteration termination condition is:

[0035] |E (k+1) - E (k) | ≤ δ

[0036] Among them, E (k) , E (k+1) are the error function values at the k-th and (k + 1)-th iterations, and δ is the convergence threshold of the error function value.

[0037] Furthermore, in step S2, constructing the roadway directed weighted topological graph according to the principle of graph theory, setting the weight of the roadway where the fire source is located to ∞ in the graph, and distinguishing the safe area, the affected area, and the high-risk area includes the following steps:

[0038] S21, obtaining the cross-connection relationship between each roadway and discarding the nodes without intersections at the turns, simplifying the topological connection relationship by the effective simplification method, setting the node set as V = (v1, v2, v3…v i ), and setting the edge set as E = (e 12 , e23 , e 34 , …e ij ), determine the connection direction between two points according to the wind direction to construct a directed topological graph G = (V, E);

[0039] S22, determine the actual length, static factor influence coefficient, and dynamic factor influence coefficient of each roadway, calculate the equivalent length of each edge, and then calculate the passing time of each edge and assign it as the weight w of the edge ij , construct a directed weighted topological graph G = (V, E, W);

[0040] The static factors include: roadway type, roadway slope, roadway height, and obstacles;

[0041] The dynamic factors include: wind speed, temperature, harmful gas concentration, and visibility;

[0042] The formula for calculating the equivalent length is:

[0043]

[0044] In the formula:

[0045] L is the equivalent length of the roadway;

[0046] K1 is the influence coefficient of roadway type;

[0047] K2 is the influence coefficient of roadway slope;

[0048] K3 is the influence coefficient of roadway height;

[0049] K4 is the influence coefficient of wind speed;

[0050] K5 is the influence coefficient of temperature;

[0051] K6 is the influence coefficient of gas concentration;

[0052] K7 is the influence coefficient of visibility;

[0053] l i is the equivalent length of obstacles in the roadway

[0054] The formula for calculating the passing time is:

[0055] Among them,

[0056] w ij is the passing time, L ij is the equivalent length, v ij is the personnel speed, set as a fixed value at the initial moment;

[0057] S23. Mark the roadway where the fire source is located in the figure and assign a weight of ∞. Divide the roadway into a safe area, an affected area, and a high-risk area according to the temperature, CO, and CO₂ concentrations in the roadway.

[0058] Further, in step S3, for monitoring the damage of the roadway and updating the topological map, at the initial moment of the fire, the method of the k-optimal path is used to plan k paths for the personnel in each area, including the following steps:

[0059] S31. Monitor the deformation of the roadway through displacement sensors and stress sensors arranged in the roadway. If either of the two conditions in the deformation threshold is exceeded, mark it as an impassable roadway, assign a weight of ∞, and update the directed weighted topological map.

[0060] The threshold conditions include: (1) The roadway deformation rate (2) The instability factor

[0061] The expression of the roadway deformation rate is:

[0062]

[0063] The expression of the instability factor is:

[0064]

[0065] Among them,

[0066] is the roadway deformation rate threshold, is the instability factor threshold, D x is the horizontal displacement, D y is the vertical displacement, W 宽 is the roadway width, σ s is the support stress, σ r is the rock mass stress, and ΔD is the displacement change rate;

[0067] S32. Starting from the personnel position and ending at the safety exit or refuge chamber, use the K-optimal path algorithm to output k escape paths for the personnel in different areas respectively.

[0068] Further, in step S4, the parameter data samples obtained from the fire simulation and the on-site parameter data samples are processed and fused, and then input into the ConvLSTM prediction model to establish a prediction model for the fire and the spread of toxic and harmful substances, so as to realize real-time dynamic disaster avoidance route planning, including the following steps:

[0069] S41. Obtain the simulation data D sim ={(x, y, z, t), T sim , C CO,sim , Vsmoke,sim , C smoke,sim , V visibility,sim}, and sensor data D sensor = {(x i , y i , z i , t i ), T sensor , C CO,sensor , V smoke,sensor , C smoke,sensor , V visibility,sensor},

[0070] S42, Data preprocessing;

[0071] S43, Build a ConvLSTM prediction model. Input the processed data X = {T t , C CO,t , V smoke,t , V visibility,t} x,y,z into the input layer in the format of a four-dimensional tensor. Extract spatial features and model temporal dependencies through the convolutional kernels of the ConvLSTM layer. Reduce the computational amount through the pooling layer. Then, map the output of the ConvLSTM layer to the final prediction result through the fully connected layer. Finally, output the prediction result for the next moment from the output layer. Define the loss function and update the model parameters through the optimizer to train the model. Establish a prediction model for fire and the spread of toxic and harmful substances, and output the data Y = {T t+1 , C CO,t+1 , V smoke,t+1 , V visibility,t+1} x,y,z .

[0072] Furthermore, in step S42, the data preprocessing includes the following steps:

[0073] S421, Calculate the visibility formula as: Then perform linear time interpolation on the data, and the formula is: Spatial interpolation processing, and the formula is:

[0074] where

[0075] V(x, y, z, t) represents the change of visibility with spatial position (x, y, z) and time t, β ext is the extinction coefficient, k is a constant, C smoke is the smoke concentration, X interp (t sim ) is the linear interpolation of time t sim , X tk is the sensor data of time t k , and t simis the interpolation time point, d i is the distance from the target to the i-th sensor, X i is the measured value of the i-th sensor;

[0076] S422. Weightedly fuse the simulation data and the interpolated sensor data. The formula is: X fused = α·X sensor +(1 - α)·X sim ,

[0077] where

[0078] X fused is the fused data value, X sensor is the real sensor data value, X sim is the simulated interpolated data value, and α is the weight factor;

[0079] S423. Standardize the data. The formula is

[0080] where

[0081] X norm is the normalized data value, X is the original variable value, X max 、X min are the maximum and minimum values of the data.

[0082] Furthermore, in step S5, the data monitored by the sensors when a fire occurs is input into the prediction model to predict the dynamic change trends of various factors affecting escape, and update the personnel escape speed and roadway weights as follows:

[0083] The monitored data is input into the prediction model to predict the change trends of various factors affecting escape, and based on the prediction data Y = {T t+1 ,C CO,t+1 ,V smoke,t+1 ,V visibility,t+1} x,y,z dynamically update the area division, roadway weights, and personnel escape speed, which specifically includes the following steps:

[0084] S51. Update the personnel escape speed according to the prediction data Y = {T t+1 ,C CO,t+1 ,V smoke,t+1 ,V visibility,t+1} x,y,z Update the personnel escape speed,

[0085] The expression for updating the personnel escape speed is:

[0086] V = V0×ω1×ω2×ω3

[0087] Among them, V0 is the normal escape speed of personnel, ω1 is the temperature influence coefficient, ω2 is the CO influence coefficient, and ω3 is the visibility influence coefficient;

[0088] S52. According to the prediction data Y = {T t+1 , C CO,t+1 , V smoke,t+1 , V visibility,t+1} x,y,z Dynamically update the roadway weights, that is, update K4, K5, K6, and K7 and substitute them into the above formula for calculating the equivalent length of the roadway. Substitute the updated equivalent length and speed into the formula for calculating the passage time to update the roadway weights;

[0089] The value formula of ω1 is:

[0090]

[0091] Among them, v m is the maximum escape speed, F1 is the actual temperature of the mine, F w1 is the temperature at which personnel feel uncomfortable, F w2 is the temperature that causes burns to the respiratory tract and the like of personnel, F d is the temperature that causes the death of personnel;

[0092] The value formula of ω2 is:

[0093]

[0094] Among them, is the CO volume fraction, and t is the time of personnel in the fire;

[0095] The value formula of ω3 is:

[0096]

[0097] Among them, Q C is the extinction coefficient, Q C = 3 / l vis , l vis is the visibility

[0098] Furthermore, in step S6, the steps of setting the escape path constraint conditions and using the ant colony algorithm to update the previous k paths and output the optimal paths in different regions are as follows:

[0099] S61. Setting the escape path constraint conditions includes: (1) Roadways where the temperature is less than or equal to 60 °C, the CO volume fraction is less than or equal to 0.0024%, and the visibility is greater than or equal to 5 m; (2) The total passage time is less than or equal to the sum of the available safe evacuation times, and the passage time of each roadway is less than or equal to the available safe evacuation time of each section of the roadway; (3) H 总累积 <H阈值 , the cumulative lethal factors consider temperature and harmful gases;

[0100] The expression for the available safe evacuation time is:

[0101] t Rest = 601.85ξexp(-0.08T)

[0102] where ξ is the roadway slope influence coefficient and T is the air temperature in the roadway;

[0103] The threshold corresponding to the cumulative lethal factor is the cumulative lethal threshold. For the calculation of the cumulative lethal threshold, first calculate the maximum tolerance time, define the maximum tolerance time as the temperature cumulative threshold, calculate the gas concentration cumulative threshold based on the semi-lethal concentration of the harmful gas and the maximum tolerance time, and calculate the comprehensive cumulative lethal threshold by considering the combined influence of temperature and gas;

[0104] The expression for the temperature cumulative threshold is:

[0105] H 温度阈值 = T max

[0106] The expression for the maximum tolerance time is:

[0107] T max = 1812exp(-0.046T)

[0108] The expression for the gas concentration cumulative threshold is:

[0109] H 气体阈值 = LC 50 ·T max

[0110] where LC 50 is the semi-lethal concentration of the gas;

[0111] The expression for the total cumulative threshold is:

[0112] H 总 = ω 温度 H 温度阈值 + ω 气体 H 气体阈值

[0113] where ω 温度 and ω 气体 are the weights of temperature and gas;

[0114] S62. Use the dynamic ant colony algorithm to update K paths in different regions. First, dynamically adjust the weights of temperature, concentration, visibility, and travel time according to the output results of the ConvLSTM prediction model. Initialize the cost function of the path as C ij (t) = ε1Tij (t) + ε2G ij (t) + ε3V ij (t) + ε4t ij + P(t), P(t) = λ H · max(0, H 总 (t) - H 总阈值 ), initialize the pheromone concentration τ ij (0) = τ0, initialize the heuristic information function as Initialize the ant positions to be randomly distributed in two regions. According to the path selection probability P ij (t) for path transfer, the expression is

[0115] where α and β are adjusted according to the output results. After each iteration, update the pheromone, and the expression is Iteratively update K paths in different regions according to the above steps;

[0116] The dynamically adjusted weights are as follows: (1) Increase the weights of ε1 and ε2 in the high-risk area; (2) In the affected area, balance and dynamically adjust the four weight coefficients; (3) Increase the weight of ε4 in the safe area;

[0117] where T ij (t), G ij (t), V ij (t), t ij are the temperature, gas concentration, visibility, and passing time on path P ij , ε1, ε2, ε3, ε4 are the weight coefficients of the corresponding factors, P(t) is the penalty term to limit the paths that do not meet the safety constraints, λ H is the penalty coefficient, H 总 (t) is the total cumulative lethal risk of the path, H 总阈值 is the lethal risk threshold, α is the pheromone importance factor, β is the heuristic function importance factor, ρ is the pheromone evaporation factor, Δτ ij (k) (t) is the pheromone increment left by the k-th ant on path P ij , Q represents the pheromone increase intensity and is a constant.

[0118] Furthermore, in step S7, the sending the optimal path information to the mobile terminal carried by the worker and the underground sign specifically means: Integrating the optimal path information including the shortest path direction and the safe haven area, sending it to the mobile terminal carried by the worker for display of the latest escape path and on the dynamic sign in the roadway, updating the evacuation guidance, and at the same time performing voice navigation on the terminal and the sign.

[0119] The beneficial effects are:

[0120] 1. The present invention proposes a real-time dynamic disaster avoidance route planning method that combines fire simulation and on-site sensor data. By fusing the parameter data samples obtained from fire simulation with the on-site monitoring data samples and inputting them into a ConvLSTM prediction model, it dynamically predicts the changes in the disaster environment, including key factors affecting escape such as fire spread and gas concentration changes. This method can update the passing speed and weight of each roadway in real time during a disaster, solving the lag problem caused by timed updates and ensuring that the system can quickly respond to the dynamic changes of the disaster.

[0121] 2. The present invention proposes a real-time monitoring and emergency response mechanism for the damage of coal mine fire roadways. By arranging displacement sensors and stress sensors in the roadways to monitor the deformation of the roadways, when the deformation or collapse of the roadway exceeds a preset threshold, the system automatically marks this area as impassable and updates the weight value in the roadway topology map. This mechanism can reflect the damage of the roadway in real time and adjust the disaster avoidance path according to the actual changes to ensure the timely response to emergencies.

[0122] 3. The present invention proposes a dynamic coal mine fire disaster avoidance path planning method. First, when a fire occurs, the k-optimal path algorithm is used to plan multiple disaster avoidance paths for personnel in different areas, providing multiple alternative routes to avoid the risk of impassability caused by a single path. Then, combined with the ConvLSTM model and on-site sensor data, the path weight and passing time are adjusted according to the fire spread trend and real-time data of each path. The path planning can be updated in real time dynamically to ensure that the optimal path is always maintained during a fire. In addition, by dynamically adjusting the weight coefficients in the path, such as temperature, gas concentration, and visibility, etc., it can preferentially optimize the paths in high-risk areas, balance the escape efficiency and safety, and quickly respond to changes through the ant colony algorithm, improving the success rate and safety of miners' escape. BRIEF DESCRIPTION OF THE DRAWINGS

[0123] Figure 1 is a flowchart of the simulation and data jointly driven coal mine fire dynamic disaster avoidance route planning method of the present invention.

[0124] Figure 2 is a flowchart of the fire source location by the spatio-temporal Transformer prediction model combined with the physical model inversion method.

[0125] Figure 3 is a flowchart of the ConvLSTM smoke spread prediction model. DETAILED DESCRIPTION OF THE INVENTION

[0126] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0127] Embodiment

[0128] This embodiment provides a method for planning a dynamic disaster avoidance route for coal mine fires driven by simulation and data jointly, and the flowchart is as Figure 1 described, including the following steps:

[0129] Step 1: Obtain data of gas concentration, temperature, and wind speed using fire sensors arranged in the roadway network, and input the processed data into a spatio-temporal Transformer prediction model combined with a physical model inversion method to locate the fire source position;

[0130] As Figure 2 shown, to locate the fire source position using a spatio-temporal Transformer prediction model combined with a physical model inversion method, first construct a spatio-temporal Transformer prediction model. After data processing, preliminarily locate the fire source position and intensity through prediction, and then correct it through a physical model. Define an error function and minimize the error function through the gradient descent algorithm, and finally output the fire source position;

[0131] Specifically, the above Step 1 includes the following steps:

[0132] Step 11: Construct a spatio-temporal Transformer model, use historical fire data to generate labeled fire source positions X true and intensities Q true as training samples, define a loss function L Transformer =||X true -X DL ||+||Q true -Q DL || and add an AdamW optimizer to establish a spatio-temporal Transformer model for preliminary fire source location;

[0133] Step 12: Perform Gaussian filtering denoising, linear time interpolation, spatial interpolation, and data normalization on the temperature T(x,y,z,t), gas concentration C(x,y,z,t), and wind speed V(x,y,z,t) data collected by the sensors to obtain a time series X={T(x,y,z,t), C(x,y,z,t), V(x,y,z,t)}. Divide the data into multiple blocks and input them into the model, use the self-attention mechanism to calculate spatial features, capture the spatial relationship between sensors, and output the preliminary fire source position XDL and the heat source intensity Q DL ;

[0134] The Gaussian filtering expression is:

[0135]

[0136] where T'(x, y) is the data after filtering, G(i, j) is the two-dimensional Gaussian kernel, σ is the standard deviation of Gaussian filtering, k is the radius of the filtering window, and exp() is the exponential function with base e;

[0137] The linear time interpolation expression is:

[0138]

[0139] where t1 and t2 are the two known time points on the left and right of the missing point, T(t1) and T(t2) are the corresponding known data values, and T(t) is the value of the missing point;

[0140] The spatial interpolation expression is:

[0141]

[0142] where X(x, y, z) is the value of the interpolation point, X i is the value of the known point i, d i is the distance between the interpolation point and the known point i, and p is the power parameter;

[0143] The data normalization expression is:

[0144]

[0145] where X is the original data, Xmin and Xmax are the maximum and minimum values of the data, and Xnorm is the normalized data;

[0146] Step 13, substitute the preliminary position X DL and the heat source intensity Q DL of the heat source into the heat conduction model and the gas diffusion model as the initial values of the physical model to correct the deep learning prediction results;

[0147] The expression of the heat conduction model is:

[0148]

[0149] where T is the temperature, in °C; α is the diffusion coefficient, in m2 / s, and Q(x, y, z) is the heat source term, in W / m3; is the Laplacian operator of temperature;

[0150] The expression of the gas diffusion model is:

[0151]

[0152] Among them, C is the gas concentration in ppm; D is the diffusion coefficient in m² / s; V is the ventilation speed in m / s; S(x, y, z) is the gas source term, representing the gas release intensity at the fire source in ppm / s; ▽ 2 C i is the Laplacian operator of the concentration, describing the diffusion behavior of the gas; v·▽C i is the convection term of the gas;

[0153] Step 14, define the error function E(X source ,Q source ) based on the sensor data and the results of the physical model, and minimize the error function minE(X source ,Q source ) by the gradient descent method to optimize the fire source location and intensity until the iteration termination condition is met, and output the finally corrected X final and Q final ;

[0154] The expression of the error function is:

[0155]

[0156] Among them, X source is the fire source location, Q source is the fire source intensity, T obs,i is the temperature observed by the sensor, T phys,i is the temperature calculated by the physical model, C obs,i is the gas concentration observed by the sensor, C phys,i is the gas concentration calculated by the physical model;

[0157] The iteration termination condition is:

[0158] |E (k+1) -E (k) |≤δ

[0159] E (k) 、E (k+1) are the error function values at the k-th and (k + 1)-th iterations, and δ is the convergence threshold of the error function value.

[0160] Step 2, construct a directed weighted topological graph of the roadway according to the principle of graph theory, mark the roadway where the fire source is located in the graph and set the weight to ∞, and distinguish between safe roadways and dangerous roadways;

[0161] Specifically, the above Step 2 includes the following steps:

[0162] Step 21: Obtain the cross - connection relationships between each roadway, discard the nodes with no cross - turns, simplify the topological connection relationships using the basic principle of the effective simplification method, set the node set as V=(v1, v2, v3…vi), the edge set as E=(e12, e23, e34,…eij), and construct a directed topological graph G=(V, E) according to the connection direction between two points determined by the wind direction. The so - called effective simplification method is to simplify the cross - connection relationships, transforming complex intersection points and connection relationships into a simplified node set and edge set;

[0163] Step 22: Determine the actual length, static - factor influence coefficient, and dynamic - factor influence coefficient of each roadway, calculate the equivalent length of each edge, and then calculate the passing time of each edge and assign it as the weight w of the edge ij , and construct a directed weighted topological graph G=(V, E, W);

[0164] The static factors include: roadway type, roadway slope, roadway height, obstacles; the dynamic factors include: wind speed, temperature, harmful gas concentration, visibility;

[0165] The expression for the equivalent length is:

[0166]

[0167] where L is the equivalent length of the roadway, K1 is the influence coefficient of roadway type, K2 is the influence coefficient of roadway slope, K3 is the influence coefficient of roadway height, K4 is the influence coefficient of wind speed, K5 is the influence coefficient of temperature, K6 is the influence coefficient of gas concentration, and K7 is the influence coefficient of visibility;

[0168] The expression for the passing time is:

[0169]

[0170] where w ij is the passing time, L ij is the equivalent length, v ij is the speed of personnel, which is set as a fixed value at the initial moment, such as v ij = 1.3m / s;

[0171] The value of K1 is as follows: The influence coefficient of roadway type K1

[0172]

[0173] The value of K2 is as follows: The influence coefficient of roadway slope K2

[0174]

[0175] The value of K3 is as follows: The influence coefficient of roadway height K3

[0176]

[0177]

[0178] The value of K4 is as follows: the wind speed influence coefficient K4

[0179]

[0180] The expression of K5 is:

[0181] The expression of K6 is:

[0182] The expression of K7 is:

[0183] The value of the local obstacle influence coefficient is as follows:

[0184] Local obstacle influence coefficient

[0185]

[0186] Step 23: Mark the roadway where the fire source is located in the figure and assign a weight of ∞. Divide the roadway into a safe area, an affected area, and a high-risk area according to the temperature, CO, and CO2 concentrations in the roadway. The affected area includes a slightly dangerous area and a moderately dangerous area;

[0187] The area division criteria are as follows:

[0188] Mine fire high-temperature hazard level

[0189]

[0190] Step 3: Monitor the roadway damage situation and update the topological map. At the initial moment of the fire, use the k-optimal path method to plan k paths for the personnel in each area;

[0191] Specifically, the above Step 3 includes the following steps:

[0192] Step 31: Monitor the roadway deformation situation through displacement sensors and stress sensors arranged in the roadway. If it exceeds the deformation threshold or collapses, mark it as an impassable roadway and assign a weight of ∞, and update the directed weighted topological map;

[0193] The threshold conditions include: (1) Roadway deformation rate (2) Instability factor

[0194] The expression of the roadway deformation rate is:

[0195]

[0196] Among them, is the threshold of roadway deformation rate, D x is the horizontal displacement, D y is the vertical displacement, W 宽 is the roadway width;

[0197] The expression of the instability factor is:

[0198]

[0199] Among them, is the instability factor threshold, σ s is the support stress, σ r is the rock mass stress, and ΔD is the displacement change rate;

[0200] Step 32: Locate the position of the personnel, mark the area where the personnel are located, and starting from the personnel position and ending at the safety exit or refuge chamber, use the K-optimal path algorithm to output k escape paths for the personnel in the safe area and the dangerous area respectively.

[0201] Step 4: In order to achieve real-time dynamic disaster avoidance route planning, the parameter data samples obtained from the fire simulation and the on-site parameter data samples are processed and fused and then input into the ConvLSTM prediction model to establish a prediction model for fire and the spread of toxic and harmful substances;

[0202] As Figure 3 shown, it is the ConvLSTM smoke spread prediction model;

[0203] Specifically, the above Step 4 includes the following steps:

[0204] Step 41: Obtain the simulation data D sim ={(x, y, z, t), T sim , C CO,sim , V smoke,sim , C smoke,sim , V visibility,sim} and the sensor data D sensor ={(x i , y i , z i , t i ), T sensor , C CO,sensor , V smoke,sensor , C smoke,sensor , V visibility,sensor};

[0205] Step 42: Data preprocessing. First, calculate the visibility, then perform linear time interpolation and spatial interpolation on the data, then weight and fuse the simulation data and the interpolated sensor data, and finally standardize the data;

[0206] The expression for calculating visibility is as follows:

[0207]

[0208] where V(x, y, z, t) is the visibility at time t at position (x, y, z), and β ext is the extinction coefficient, k is the proportionality constant, and C smoke is the smoke concentration;

[0209] The linear time interpolation expression is as follows:

[0210]

[0211] where X interp (t sim ) is the value of the variable after interpolation and then simulation at time t sim value, X tk is the variable value at time point t k time, t k+1 is the variable value at time point t k+1 time, t sim is the target interpolation time point, t k and t k+1 are the two endpoints of the time interval used for interpolation;

[0212] The spatial interpolation processing expression is as follows:

[0213]

[0214] where X interp (x, y, z) is the variable value after interpolation at the spatial position (x, y, z), X i is the known variable value of the spatial interpolation point i, d i is the distance between the interpolation point and the target position (x, y, z), and p is the weight exponent that controls the rate of attenuation of the interpolation weight with distance;

[0215] The weighted fusion expression of the simulation data and the interpolated sensor data is: X fused = α·X sensor + (1 - α)·X sim where X fused is the fused value, α is the weight coefficient, X sensor is the sensor observation value, and X sim is the simulation calculation value;

[0216] The normalization processing expression is as follows:

[0217]

[0218] where X normThe standardized variable value, the original variable value of X, X min The minimum value of the variable, X max The maximum value of the variable;

[0219] Step 43, construct a ConvLSTM prediction model, and input the processed data X = {T t , C CO,t , V smoke,t , V visibility,t} x,y,z into the input layer in the format of a four-dimensional tensor, extract spatial features and model time dependencies through the convolutional kernels of the ConvLSTM layer, reduce the computational amount through the pooling layer, then map the output of the ConvLSTM layer to the final prediction result by the fully connected layer, and finally output the prediction result of the next moment by the output layer. Define the loss function and update the model parameters through the optimizer to train the model, establish a prediction model for the spread of fire and toxic and harmful substances, and output the data of the next moment Y = {T t+1 , C CO,t+1 , V smoke,t+1 , V visibility,t+1} x,y,z .

[0220] Step 5, input the data monitored by the sensor when a fire occurs into the prediction model, predict the dynamic change trend of each factor affecting escape, and update the personnel escape speed and roadway weight;

[0221] Specifically, the above step 5 includes the following steps:

[0222] Step 51, according to the output data of the next moment Y = {T t+1 , C CO,t+1 , V smoke,t+1 , V visibility,t+1} x,y,z update the personnel escape speed,

[0223] The expression for updating the personnel escape speed is:

[0224] V = V0 × ω1 × ω2 × ω3

[0225] where V0 is the normal escape speed of personnel, ω1 is the temperature influence coefficient, ω2 is the CO influence coefficient, and ω3 is the visibility influence coefficient;

[0226] Step 52, according to the output data of the next moment Y = {T t+1 , C CO,t+1 , V smoke,t+1 , V visibility,t+1} x,y,zUpdate the roadway weight, that is, update K4, K5, K6, and K7 and substitute them into the above calculation formula of the equivalent length of the roadway. Substitute the updated equivalent length and speed into the travel time calculation formula to update the roadway weight;

[0227] The value formula of ω1 is:

[0228]

[0229] Among them, v m is the maximum escape speed, and the value is determined according to the actual situation; F1 is the actual temperature of the mine; F w1 is the temperature at which people feel uncomfortable, taking 30 °C; F w2 is the temperature that causes burns to the respiratory tract and the like of people, taking 60 °C; F d is the temperature that causes people to die, taking 120 °C;

[0230] The value formula of ω2 is:

[0231]

[0232] Among them, is the CO volume fraction, and t is the time of people in the fire;

[0233] The value formula of ω3 is:

[0234]

[0235] Among them, Q C is the light extinction coefficient, Q C = 3 / l vis , l vis is the visibility

[0236] Step 6, set the escape path constraint conditions and use the ant colony algorithm to update the previous k paths, and output the optimal paths in different regions;

[0237] Specifically, the above step 6 includes the following steps:

[0238] Step 61, set the escape path constraint conditions including: (1) roadways with a temperature less than or equal to 60 °C, a CO volume fraction less than or equal to 0.0024%, and a visibility greater than or equal to 5 m; (2) the total travel time is less than or equal to the sum of the available safe evacuation times, and the travel time of each roadway is less than or equal to the available safe evacuation time of each section of the roadway; (3) H total cumulative < H threshold, and the cumulative lethal factors mainly consider temperature and harmful gases;

[0239] The expression of the safe evacuation time is:

[0240] t Rest = 601.85ξexp(-0.08T)

[0241] Among them, ξ is the influence coefficient of roadway gradient, and T is the air temperature in the roadway;

[0242] The threshold value corresponding to the cumulative lethal factor is the cumulative lethal threshold. For the calculation of the cumulative lethal threshold, first calculate the maximum tolerance time, define the maximum tolerance time as the temperature cumulative threshold, calculate the gas concentration cumulative threshold according to the semi-lethal concentration of harmful gas and the maximum tolerance time, and calculate the comprehensive cumulative lethal threshold by comprehensively considering the influences of temperature and gas;

[0243] The expression of the temperature cumulative threshold is:

[0244] H 温度阈值 =T max

[0245] The expression of the maximum tolerance time is:

[0246] T max =1812exp(-0.046T)

[0247] The expression of the gas concentration cumulative threshold is:

[0248] H 气体阈值 =LC 50 ·T max

[0249] Among them, LC 50 is the semi-lethal concentration of the gas;

[0250] The expression of the total cumulative threshold is:

[0251] H 总 =ω 温度 H 温度阈值 +ω 气体 H 气体阈值

[0252] Among them, ω 温度 and ω 气体 are the weights of temperature and gas;

[0253] Step 62: Use the dynamic ant colony algorithm to update the K paths in different regions. First, dynamically adjust the weights of temperature, concentration, visibility, and travel time according to the output results of the ConvLSTM prediction model, initialize the cost function of the path, initialize the pheromone concentration, initialize the heuristic information function, initialize the random distribution of ant positions in two regions, perform path transfer according to the path selection probability, make adjustments according to the output results, update the pheromone after each round of iteration, and perform iterative updates according to the above steps to output the K paths in different regions;

[0254] The dynamic weight adjustment is as follows: 1) In high-risk areas, increase the weights of ε1, ε2, and ε3, prioritize safety, and maximize the risk avoidance ability; 2) In affected areas, balance and dynamically adjust the four weight coefficients to achieve a balance; 3) In safe areas, increase the weight of ε4, prioritize speed, and minimize the evacuation time.

[0255] The cost function expression of the path is:

[0256] C ij (t) = ε1T ij (t) + ε2G ij (t) + ε3V ij (t) + ε4t ij + P(t), P(t) = λ H ·max(0, H 总 (t) - H 总阈值 )

[0257] Where, T ij (t), G ij (t), V ij (t), t ij are the temperature, gas concentration, visibility, and travel time on path P ij , ε1, ε2, ε3, ε4 are the weight coefficients of the corresponding factors, P(t) is the penalty term to limit the path that does not meet the safety constraints, λ H is the penalty coefficient, H 总 (t) is the total cumulative lethal risk of the path, H 总阈值 is the cumulative lethal risk threshold;

[0258] The expression of the heuristic information function is:

[0259]

[0260] In one embodiment, the expression of the path selection probability is:

[0261]

[0262] Where, α is the pheromone importance factor, and β is the heuristic function importance factor;

[0263] The expression of pheromone update is:

[0264]

[0265] Where, ρ is the pheromone evaporation factor, Δτ ij (k) (t) is the pheromone increment left by the kth ant on path P ij , Q represents the pheromone increase intensity and is a constant.

[0266] Step 7: Send the shortest path information to the mobile terminal carried by the worker and the underground signpost. The optimal path information updated according to Step 6 includes the shortest path direction and the safe haven area, which are sent to the mobile terminal carried by the worker to display the latest escape path and updated evacuation guidance on the dynamic signpost in the roadway. At the same time, voice navigation is carried out on the terminal and the signpost.

[0267] Finally, it should be noted that the above are only the preferred embodiments of the present invention and are not used to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, for those skilled in the art, they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for dynamic mine fire avoidance route planning driven by simulation and data, characterized in that: The following steps are involved: S1: Use fire sensors arranged in the tunnel network to obtain data on gas concentration, temperature and wind speed. After data processing, use the spatiotemporal Transformer prediction model combined with the physical model inversion method to locate the fire source; S2: According to the principle of graph theory, a directed weighted topological graph of the lanes is constructed, the lanes where the fire source is located are marked in the graph, the weight is set to ∞, and the safe area, the affected area, and the high-risk area are distinguished, and the affected area includes the light-risk area and the medium-risk area; S3: Monitor the damage to the tunnels and update the topology map. At the initial moment of the fire, use the k-optimal path method to plan k paths for personnel in each area. S4: After processing and fusion of parameter data samples obtained from fire simulation and on-site parameter data samples, they are input into the ConvLSTM prediction model to establish a model for predicting the spread of fire and toxic and hazardous substances, so as to realize real-time dynamic disaster avoidance route planning; S5: The data monitored by the sensor when a fire occurs is input into the prediction model to predict the changing trends of various factors affecting escape, and update the escape speed of personnel and the weight of the lane; S6: Set the escape path constraints, and use the ant colony algorithm to update the previous k paths, and output the optimal paths in different areas; S7: Send the optimal path information to the mobile terminal carried by the workers and the underground signboard.

2. The method for dynamic mine fire avoidance route planning driven by simulation and data according to claim 1 is characterized in that: In step S1, the fire sensors arranged in the tunnel network are used to obtain data on gas concentration, temperature and wind speed, and after data processing, the fire source is located using a spatiotemporal Transformer prediction model combined with a physical model inversion method, including the following steps: S11, build a spatiotemporal Transformer model, use historical fire data to generate labeled fire source locations and intensities as training samples, and define the loss function L Transformer =||X true -X DL ||+||Q true -Q DL ||And add AdamW optimizer to build a spatiotemporal Transformer model for preliminary fire source location; S12, the temperature T(x,y,z,t), gas concentration C(x,y,z,t), wind speed V(x,y,z,t) data collected by the sensor are subjected to Gaussian filtering denoising, linear time interpolation, spatial interpolation and data normalization to obtain the time series X = {T(x,y,z,t), C(x,y,z,t), V(x,y,z,t)}, the data is divided into multiple blocks and input into the model, the spatial features are calculated using the self-attention mechanism, the spatial relationship between sensors is captured, and the preliminary location of the fire source X is output. DL and fire intensity Q DL ; The Gaussian filtering formula is: The linear time interpolation formula is The spatial interpolation formula is: The data normalization formula is: in, exp() is an exponential function with base e, T'(x,y) is the filtered data, G(i,j) is a two-dimensional Gaussian kernel, σ is the standard deviation of the Gaussian filter, k is the radius of the filter window, t1 and t2 are two known time points on the left and right of the actual point, T(t1) and T(t2) are the corresponding known data values, T(t) is the missing point value, X(x,y,z) is the value of the interpolated point, X i The value of point i is known, d i is the distance between the interpolation point and the known point i, p is the power parameter, X is the original data, X min , X max is the maximum and minimum value of the data, X norm is the normalized data; S13, the initial location of the fire source X DL and fire intensity Q DL Substitute the initial values ​​of the physical model into the heat conduction model and gas diffusion model to correct the deep learning prediction results; The expression of the heat conduction model is: The gas diffusion model expression is: in, C is the gas concentration, D is the diffusion coefficient, V is the ventilation speed, S(x,y,z) is the gas source term, which indicates the gas release intensity at the fire source, T is the temperature, α is the diffusion coefficient, Q(x,y,z) is the heat source term, is the Laplace operator of concentration, describing the diffusion behavior of the gas, is the convection term of the gas, is the temperature Laplace operator; S14, define the error function based on sensor data and physical model results Minimize the error function minE(X source ,Q source )Optimize the fire location and intensity until |E (k +1) -E (k) |≤δ terminates the iteration and outputs the final corrected X final and Q final ; in, X source is the fire source location, Q source is the fire intensity, T obs,i is the temperature observed by the sensor, T phys,i Calculate the temperature for the physical model, C obs,i is the gas concentration observed by the sensor, C phys,i Calculate gas concentrations for physical models, The iteration termination condition is: |And (k+1) -AND (k) |≤δ Among them, E (k) 、E (k+1) is the error function value at the kth and k+1th iterations, and δ is the convergence threshold of the error function value.

3. According to the method for dynamic coal mine fire avoidance route planning driven by simulation and data combination according to claim 1, it is characterized in that: In step S2, the method of constructing a directed weighted topological graph of the lanes according to the principle of graph theory, marking the lanes where the fire source is located in the graph with the weight set to ∞, and distinguishing the safe area, the affected area, and the high-risk area includes the following steps: S21, obtain the cross-connection relationship between the lanes and discard the nodes without intersections, use the effective simplification method to simplify the topological connection relationship, and set the node set to V = (v1, v2, v3...v i ), the edge set is set to E = (e 12 ,e 23 ,e 34 ,…e ij ), determine the connection direction between two points according to the wind direction and construct a directed topological graph G = (V, E); S22, determine the actual length of each lane, the static factor influence coefficient and the dynamic factor influence coefficient, calculate the equivalent length of each edge, and then calculate the travel time of each edge and assign it as the edge weight w ij , construct a directed weighted topological graph G = (V, E, W); The static factors include: roadway type, roadway slope, roadway height and obstacles; The dynamic factors include: wind speed, temperature, concentration of harmful gases and visibility; The calculation formula of the equivalent length is: Where: L is the equivalent length of the tunnel; K1 is the influence coefficient of the roadway type; K2 is the influence coefficient of roadway slope; K3 is the influence coefficient of tunnel height; K4 is the wind speed influence coefficient; K5 is the temperature influence coefficient; K6 is the gas concentration influence coefficient; K7 is the visibility influence coefficient; l i is the equivalent length of the obstacle in the tunnel The travel time calculation formula is: in, w ij is the travel time, L ij is the equivalent length, v ij is the personnel speed, which is set to a fixed value at the initial moment; S23, mark the lane where the fire source is located in the figure and assign the weight to ∞, and divide the lane into safe area, affected area and high-risk area according to the temperature, CO and CO2 concentration in the lane.

4. The method for dynamic mine fire avoidance route planning driven by simulation and data combination according to claim 1 is characterized in that: In step S3, the monitoring of the tunnel damage situation updates the topological map, and at the initial moment of the fire, the method of using k optimal paths to plan k paths for personnel in each area includes the following steps: S31, monitoring the deformation of the lane through displacement sensors and stress sensors arranged in the lane, if any of the two conditions in the deformation threshold is exceeded, marking it as an impassable lane, assigning a weight of ∞, and updating the directed weighted topological map; The threshold conditions include: (1) tunnel deformation rate (2) Instability Factor The expression of roadway deformation rate is: The expression of the instability factor is: in, is the tunnel deformation rate threshold, is the instability factor threshold, D x is the horizontal displacement, D y is the vertical displacement, W 宽 is the lane width, σ s is the support stress, σ r is the rock mass stress, ΔD is the displacement change rate; S32, taking the personnel position as the starting point and the safety exit or refuge chamber as the end point, using the K optimal path algorithm, output k escape paths for personnel in different areas respectively.

5. The method for dynamic mine fire avoidance route planning driven by simulation and data combination according to claim 1 is characterized in that: In step S4, the parameter data samples obtained by fire simulation and the on-site parameter data samples are processed and fused and input into the ConvLSTM prediction model to establish a model for predicting the spread of fire and toxic and hazardous substances to achieve real-time dynamic disaster avoidance route planning, including the following steps: S41, obtain simulation data D sim ={(x,y,z,t),T sim ,C CO,sim ,V smoke,sim ,C smoke,sim ,V visibility,sim } and sensor data D sensor ={(x i ,y i ,z i ,t i ),T sensor ,C CO,sensor ,V smoke,sensor ,C smoke,sensor ,V visibility,sensor }, S42, data preprocessing; S43, build a ConvLSTM prediction model, and convert the processed data X = {T t ,C CO,t ,V smoke,t ,V visibility,t } x,y,z The data is input into the input layer in the form of a four-dimensional tensor. The spatial features are extracted and the temporal dependency is modeled through the convolution kernel of the ConvLSTM layer. The amount of calculation is reduced through the pooling layer. The output of the ConvLSTM layer is then mapped to the final prediction result by the fully connected layer. Finally, the prediction result of the next moment is output by the output layer. The loss function is defined and the model parameters are updated through the optimizer to train the model. A model for predicting the spread of fire and toxic and hazardous substances is established, and the data Y = {T t+1 ,C CO,t+1 ,V smoke,t+1 ,V visibility,t+1 } x,y,z .

6. The method for dynamic mine fire avoidance route planning driven by simulation and data according to claim 5 is characterized in that: In step S42, the data preprocessing includes the following steps: S421, the formula for calculating visibility is: Then the data is interpolated linearly in time, the formula is: Spatial interpolation processing, the formula is: in, V(x,y,z,t) represents the change of visibility with spatial position (x,y,z) and time t, β ext is the extinction coefficient, k is a constant, C smoke is the smoke density, X interp (t sim ) is time t sim Linear interpolation of X tk is time t k Sensor data, t sim is the interpolation time point, d i is the distance from the target to the i-th sensor, X i is the measurement value of the i-th sensor; S422, weighted fusion of simulation data and interpolated sensor data, the formula is: X fused =α·X sensor +(1-α)·X sim , in, X fused is the fused data value, X sensor is the real sensor data value, X sim is the simulated interpolation data value, α is the weight factor; S423, standardize the data, the formula is in, X norm is the normalized data value, X is the original variable value, X max , X min is the maximum and minimum value of the data.

7. The method for dynamic mine fire avoidance route planning driven by simulation and data combination according to claim 1 is characterized in that: In step S5, the data monitored by the sensor when the fire occurs is input into the prediction model to predict the changing trend of various factors affecting escape, and update the escape speed of personnel and the weight of the lane as follows: The monitored data is input into the prediction model to predict the changing trend of various factors affecting escape, and according to the prediction data Y = {T t+1 ,C CO,t+1 ,V smoke,t+1 ,V visibility,t+1 } x,y,z Dynamically update the area division, lane weight and personnel escape speed, including the following steps: S51, according to the predicted data Y={T t+1 ,C CO,t+1 ,V smoke,t+1 ,V visibility,t+1 } x,y,z Update personnel escape speed, Update the personnel escape speed expression: V=V0×ω1×ω2×ω3 Among them, V0 is the normal escape speed of personnel, ω1 is the temperature influence coefficient, ω2 is the CO influence coefficient, and ω3 is the visibility influence coefficient; S52, according to the predicted data Y={T t+1 ,C CO,t+1 ,V smoke,t+1 ,V visibility,t+1 } x,y,z Dynamically update the lane weight, that is, update K4, K5, K6, K7 and substitute them into the above lane equivalent length calculation formula, substitute the updated equivalent length and speed into the travel time calculation formula to update the lane weight; The formula for determining the value of ω1 is: Among them, v m is the maximum escape speed, F1 is the actual temperature of the mine, F w1 For personnel to feel uncomfortable temperature, F w2 F is the temperature that causes burns to the respiratory tract, etc. d The temperature that caused the death; The formula for determining the value of ω2 is: in, is the CO volume fraction, t is the time the personnel are in the fire; The formula for determining the value of ω3 is: Among them, Q C is the dimming coefficient, Q C =3 / l vis ,l vis For visibility.

8. The method for dynamic mine fire evacuation route planning driven by simulation and data according to claim 1 is characterized in that: In step S6, the setting of escape path constraints, and the use of the ant colony algorithm to update the previous k paths and output the optimal paths in different areas include the following steps: S61, set the escape path constraint conditions including: (1) the temperature is less than or equal to 60℃, the volume fraction of CO is less than or equal to 0.0024%, and the visibility is greater than or equal to 5m; (2) the total travel time is less than or equal to the sum of the available safe evacuation time, and the travel time of each lane is less than or equal to the available safe evacuation time of each lane; (3) H 总累积 <H 阈值 , the cumulative lethal factors take into account temperature and harmful gases; The available safe evacuation time expression is: t Rest =601.85ξexp(-0.08T) Among them, ξ is the influence coefficient of the roadway slope, and T is the air temperature in the roadway; The threshold corresponding to the cumulative lethal factor is the cumulative lethal threshold. The calculation of the cumulative lethal threshold first calculates the maximum tolerance time, defines the maximum tolerance time as the temperature cumulative threshold, calculates the gas concentration cumulative threshold according to the half-lethal concentration of the harmful gas and the maximum tolerance time, and calculates the comprehensive cumulative lethal threshold by comprehensively considering the influence of temperature and gas; The temperature accumulation threshold expression is: H 温度阈值 =T max The maximum tolerance time expression is: T max =1812exp(-0.046T) The gas concentration accumulation threshold expression is: H 气体阈值 =LC 50 ·T max Among them, LC 50 is the half-lethal concentration of gas; The expression for the total cumulative threshold is: H 总 =ω 温度 H 温度阈值 +ω 气体 H 气体阈值 Among them, ω 温度 ,ω 气体 are the temperature and gas weights; S62, using the dynamic ant colony algorithm to update the K paths in different areas, first dynamically adjust the weights of temperature, concentration, visibility and travel time according to the output results of the ConvLSTM prediction model, and initialize the cost function of the path to C ij (t) = ε1T ij (t)+ε2G ij (t)+ε3V ij (t)+ε4t ij +P(t),P(t)=λ H max(0,H 总 (t)-H 总阈值 ), initialize the pheromone concentration τ ij (0) = τ0, the initialization heuristic information function is Initialize the ant positions randomly distributed in two areas, and select the path according to the probability P ij (t) is used for path transfer, and the expression is: Among them, α and β are adjusted according to the output results, and the pheromone is updated after each round of iteration. The expression is: According to the above steps, iteratively update K paths in different areas; The dynamic adjustment weights are: (1) the high-risk area increases the weights of ε1, ε2, and ε4; (2) the affected area dynamically adjusts the four weight coefficients in a balanced manner; (3) the safe area increases the weight of ε4; Among them, T ij (t), G ij (t), V ij (t), t ij For path P ij The temperature, gas concentration, visibility and travel time on the road, ε1, ε2, ε3, ε4 are the weight coefficients of the corresponding factors, P(t) is the penalty term that restricts the path that does not meet the safety constraints, λ H is the penalty coefficient, H 总 (t) is the total cumulative lethal risk of the path, H 总阈值 is the lethal risk threshold, α is the pheromone importance factor, β is the heuristic function importance factor, ρ is the pheromone volatility factor, Δτ ij (k) (t) is the kth ant on path P ij The pheromone increment left on the surface, Q represents the pheromone increase intensity, which is a constant.

9. The method for dynamic mine fire avoidance route planning driven by simulation and data combination according to claim 1 is characterized in that: In step S7, the sending of the optimal path information to the mobile terminal carried by the workers and the underground signboards is specifically: integrating the optimal path information including the shortest path direction and the safe haven area, sending it to the mobile terminal carried by the workers to display the latest escape path and the dynamic signboards in the tunnel, updating the evacuation instructions, and performing voice navigation on the terminal and the signboards.

Images