Method for preventing and extinguishing fire in an internal combustion engine

Abstract

This invention belongs to the field of locomotive fire prevention technology, specifically relating to a fire prevention and extinguishing method for diesel locomotives. The method includes real-time acquisition of multi-source environmental signals from the protected area of ​​the diesel locomotive, feature fusion of the acquired signals, and outputting fire identification results. Based on the fire identification results, a pre-trained cellular automata model is used to predict the fire spread trend. By determining the initial parameters of the fire source, discretizing the three-dimensional cellular mesh of the protected area, calculating the fire spread velocity vector and cellular state transition probability, and combining multiple rounds of iterative evolution calculations, accurate predictions of the fire spread range, radius of the danger zone, volume of the danger zone, and spread direction are achieved. Simultaneously, by constructing a state transition rule base, defining cellular multi-states, and performing thermodynamic calculations such as heat radiation flux and thermal balance, the prediction results are made to closely match the actual environmental characteristics of the diesel locomotive protected area, overcoming the limitations of traditional fire extinguishing technologies that rely solely on passive extinguishing without prior prediction.

Description

Technical Field

[0001] This invention belongs to the field of locomotive fire prevention technology, specifically relating to a method for fire prevention and extinguishing of internal combustion locomotives. Background Technology

[0002] Currently, fire prevention and control technologies for internal combustion locomotives mainly rely on traditional point-based temperature detection and manual fire extinguishing methods, which suffer from prominent problems such as slow response and low fire extinguishing efficiency. Existing fire detection systems mostly use single sensors or simple threshold judgments, making it difficult to achieve accurate identification under complex operating conditions; while fixed fire extinguishing devices usually use a uniform spraying method, which cannot perform targeted and precise spraying based on the location and spread trend of the fire source, resulting in serious waste of extinguishing agents and poor suppression effect. Summary of the Invention

[0003] To address the aforementioned shortcomings of existing technologies, this invention provides a method for fire prevention and extinguishing in internal combustion locomotives, comprising: S1. Real-time collection of multi-source environmental signals in the protection area of ​​the diesel locomotive using point-type ultraviolet flame detectors, heat detectors, and smoke detectors; S2. Using a deep learning algorithm based on convolutional neural networks and attention mechanisms, feature fusion is performed on the multi-source environmental signals collected in step S1, and the fire identification result is output. S3. Based on the identification results of step S2, use the pre-trained cellular automata model to predict the spread trend of the fire within the protection zone of the internal combustion locomotive. S4. Based on the fire identification results of step S2 and the fire spread trend prediction results of step S3, optimize and generate fire extinguishing instructions that include extinguishing dosage and spray angle. S5. Based on the fire extinguishing command generated in step S4, control the dosage release and the spray angle of the fixed thermal aerosol fire extinguishing device installed inside the diesel locomotive to execute the fire extinguishing action.

[0004] Further improvements to this technical solution include step S2, which includes: S21. Preprocess the multi-source environmental signals collected in step S1 to obtain a standardized signal sequence; S22. Input the standardized signal sequence obtained in step S21 into a one-dimensional convolutional neural network, and extract the time-frequency features of each detector signal through multi-layer convolution operations to generate ultraviolet feature matrix, temperature feature matrix and smoke feature matrix. S23. Input the three feature matrices generated in step S22 into the multi-head attention mechanism module, obtain the attention weight by calculating the similarity between the query matrix and the key matrix, perform weighted fusion on the value matrix, and output the fused feature vector. S24. Input the fused feature vector output in step S23 into the fully connected layer, calculate the fire identification probability through the Sigmoid activation function, and output the fire identification result when the fire identification probability exceeds the preset threshold.

[0005] Further improvements to this technical solution include step S3, which includes: S31. Determine the initial parameters of the fire source based on the fire identification results output in step S2, including the fire source location coordinates, initial heat release rate and initial fire source radius, and initialize the cellular state matrix of the cellular automata model. S32. Discretize the protection area of ​​the internal combustion locomotive into a three-dimensional cellular mesh. Each cell records the temperature state, combustion state and fuel density state, and establishes a mapping relationship between the cell state and the initial parameters of the fire source in step S31. S33. Based on the initial heat release rate and environmental parameters of the protected area determined in step S31, calculate the fire spread velocity vector and determine the transition probability of the cell state from the safe state to the heated state. S34. Update the cell state matrix based on the transition probability calculated in step S33, iteratively calculate the cell temperature change and combustion state expansion at each time step, and predict the fire spread range at multiple future time steps. S35. Based on the cell state matrix updated in step S34, extract the boundary envelope of the open flame cell set, calculate the radius and volume of the hazardous area, and output the fire spread trend prediction results.

[0006] A further improvement to this technical solution is that the formula for calculating the initial heat release rate of the fire source in step S31 is: ; In the formula: The initial heat release rate of the fire source; The preset maximum heat release rate; The fire identification probability is output in step S2. The fire source growth coefficient; The formula for calculating the fire spread rate in step S33 is: ; In the formula: This is the fire spread velocity vector; Based on the spread rate; The radiative heat feedback coefficient; Let t be the heat release rate of the fire source. The thermal conductivity of air; The ambient temperature; Area of ​​the fire source; This represents the ambient airflow velocity vector. The unit vector is the direction of propagation; The formula for calculating the cell state transition probability in step S34 is: ; In the formula: The cell state transition probability; Let Δt be the cell state at time t, where 0 represents the safe state and 1 represents the heated state; Δt is the time step; and Δx is the cell size. The formula for calculating the radius of the hazardous area in step S35 is: ; In the formula: For the predicted time The radius of the danger zone; The initial fire source radius determined in step S31; For prediction in the time domain; For integration time; is the turbulent diffusion coefficient.

[0007] Further improvements to this technical solution include step S34, which includes: S341. Construct a state transition rule base for the cellular automaton model, define four basic states of the cell including safe state, heated state, open flame state and extinguished state, and establish a mapping relationship between the state transition conditions and the transition probabilities calculated in step S33. S342. Initialize the cellular space topology, divide the protected area into a regular cubic cell array, assign fuel property parameters and thermodynamic property parameters to the cells, and construct the adjacency relationship matrix between cells. S343. Based on the state transition rule base defined in step S341 and the cellular space topology constructed in step S342, perform iterative evolution calculations. At each time step, update the state variables of all cells according to the transition probability calculated in step S33, and record the spatiotemporal evolution process of the cell temperature field and the combustion front. S344. Post-process the spatiotemporal sequence of the cellular state obtained by the iterative evolution in step S343, extract the key feature parameters of fire spread, including spread speed, spread direction and dangerous area boundary, and generate the fire spread trend prediction result required in step S4.

[0008] Further improvements to this technical solution include step S343, which includes: S3431. At the start of the current time step, traverse all cells, check the transition conditions of each cell according to the state transition rule base defined in step S341, and calculate the heat radiation flux of the adjacent open flame cell to the cell. S3432. Based on the heat radiation flux calculated in step S3431 and the current cell temperature, determine whether the state transition conditions are met. If the cell temperature exceeds the ignition temperature and the oxygen concentration is higher than the critical value, then update the cell state from the heated state to the open flame state. S3433. Perform thermal balance calculations on all cells updated in step S3432, update the temperature field based on heat conduction and heat convection exchange between cells, and advance the time step to the next iteration cycle. S3434. Repeat steps S3431 to S3433 until the preset prediction time domain is reached, and output complete cell state spatiotemporal evolution data to step S344.

[0009] A further improvement to this technical solution is that the formula for calculating the thermal radiation flux in step S3431 is as follows: ; In the formula: This represents the thermal radiation flux received by the current cell. The set of adjacent open flame cells is determined by the adjacency matrix in step S342; Let be the heat release rate of the m-th adjacent open flame cell; For the perspective coefficient; The distance between the center of the current cell and the m-th adjacent open flame cell; The state transition judgment formula for step S3432 is: ; In the formula: The updated cell state is represented by 2 for open flame, 1 for heated state, and 0 for safe state. This refers to the fuel ignition temperature. The critical oxygen volume fraction; The cell size defined in step S342; The heat balance calculation formula for step S3433 is as follows: ; In the formula: This represents the total heat received by the cell. The surface area of ​​the cell; Equivalent thermal conductivity; The temperature of the adjacent cell n; This represents the contact area between cells; The convective heat transfer coefficient; The temperature of the air surrounding the cell.

[0010] Further improvements to this technical solution include step S4, which includes: S41. Determine the fire intensity level based on the fire identification results output in step S2, and calculate the adaptive extinguishing dose by combining the dangerous area volume in the fire spread trend prediction results in step S3. S42. Based on the fire spread trend prediction results in step S3, determine the fire extinguishing activation timing and generate a fire extinguishing sequence command containing the activation time. S43. Based on the fire source location coordinates and spread direction in the fire spread trend prediction results predicted in step S3, and combined with the airflow field distribution in the diesel locomotive protection area, calculate the optimal injection angle of the directional guide nozzle. S44. The adaptive extinguishing dosage calculated in step S41, the extinguishing timing command determined in step S42, and the optimal spray angle calculated in step S43 are encapsulated into commands to generate a complete extinguishing command set and transmit it to the fixed thermal aerosol extinguishing device.

[0011] A further improvement to this technical solution is that the adaptive fire extinguishing dosage calculation formula in step S41 is as follows: ; In the formula: For adaptive extinguishing dosage; The baseline extinguishing dose is γ; γ is the fire intensity correction factor. The fire intensity level is the fire identification probability output from step S2. The values ​​obtained through mapping range from 1 to 5; It is a non-linear exponent; The predicted time for step S3 The volume of the danger zone; For reference, the volume of the protected area is used. This is the leakage compensation coefficient; For prediction and correction coefficients; The formula for calculating the fire extinguishing activation time in step S42 is: ; In the formula: This is the moment when fire suppression is initiated; The current moment; Delay time for automatic mode; The threshold for automatic startup; This is a manually triggered time. To predict the start-up time, when the radius of the danger zone predicted in step S3... Exceeding the threshold Start in advance; This is the threshold for the danger radius; The formula for calculating the optimal injection angle in step S43 is: ; ; In the formula: The optimal horizontal deflection angle; To achieve the optimal vertical pitch angle; The coordinates of the fire source location predicted in step S3; To determine the coordinates of the guide nozzle installation position; This is the horizontal angle correction amount based on computational fluid dynamics simulation; This is the vertical angle correction amount based on computational fluid dynamics simulation; This is the correction amount for the predicted level based on the spread prediction in step S3; This is the predicted vertical correction amount based on the spread prediction in step S3.

[0012] Further improvements to this technical solution include step S5, which includes: S51. Analyze the fire extinguishing command generated in step S4, extract the adaptive fire extinguishing dosage, optimal spray angle and start-up time control parameters, and verify the completeness and effectiveness of the command; S52. Based on the optimal injection angle analyzed in step S51, the electric adjustment mechanism is driven by the proportional-integral-derivative control algorithm to adjust the horizontal deflection angle and vertical pitch angle of the directional guide nozzle to the target position. S53. After the adjustment in step S52 is completed, the fixed thermal aerosol fire extinguishing device is triggered according to the start time analyzed in step S51, and the graded dose release is performed. The first stage releases large-diameter aerosol particles to quickly suppress open flames, and the second stage releases small-diameter aerosol particles to maintain the suspension concentration. S54. Monitor the actual release dose and nozzle angle feedback during the execution of step S53, compare them in real time with the target parameters analyzed in step S51, generate an execution deviation correction signal, and ensure that the fire extinguishing action is executed accurately.

[0013] The beneficial effects of this invention are as follows: This method acquires multi-source environmental signals using point-type ultraviolet flame detectors, heat detectors, and smoke detectors, overcoming the limitations of traditional single-sensor or simple threshold judgments. It combines feature fusion with a deep learning algorithm based on convolutional neural networks and attention mechanisms. Through multi-step processing including standardized preprocessing, time-frequency feature extraction, multi-head attention weighted fusion, and Sigmoid activation function probability determination, it effectively mines the correlation features of signals from different sensors, accurately distinguishing fire signals from interference signals under complex operating conditions of diesel locomotives, thus achieving accurate fire identification. Simultaneously, the real-time acquisition of multi-source signals and the efficient computation of deep learning algorithms significantly shorten the fire identification response time, solving the problem of response lag in traditional detection technologies and significantly reducing the false alarm rate under complex operating conditions, providing a reliable basis for the timely initiation of subsequent firefighting actions.

[0014] This method, based on fire identification results, uses a pre-trained cellular automata model to predict fire spread trends. By determining initial fire source parameters, discretizing the three-dimensional cellular mesh of the protected area, calculating the fire spread velocity vector and cellular state transition probabilities, and combining multi-round iterative evolution calculations, it achieves accurate predictions of the fire spread range, hazardous area radius, hazardous area volume, and spread direction. Simultaneously, by constructing a state transition rule base, defining cellular multi-states, and performing thermodynamic calculations such as thermal radiation flux and thermal balance, the prediction results closely match the actual environmental characteristics of the diesel locomotive protected area. This overcomes the limitations of traditional fire extinguishing technologies that lack advance prediction and rely solely on passive fire suppression. It enables early understanding of fire development patterns, providing scientific data support for optimizing extinguishing dosage, spray angle, and activation timing, thus achieving a technological upgrade from passive fire suppression to proactive prevention and control.

[0015] This method calculates the required extinguishing agent dosage based on the fire intensity level determined by fire identification results and the volume of the hazardous area predicted by the fire spread trend, using an adaptive extinguishing agent dosage formula. It also incorporates parameters such as leakage compensation coefficient and prediction correction coefficient to match the extinguishing agent dosage with the actual fire intensity, the hazardous area range, and the airflow characteristics of the diesel locomotive protection area. Furthermore, through a tiered dosage release mode, the first stage releases large-diameter aerosol particles to rapidly suppress open flames, while the second stage releases small-diameter aerosol particles to maintain suspension concentration. This achieves on-demand distribution and efficient utilization of the extinguishing agent, solving the problem of severe extinguishing agent waste caused by the uniform spraying of traditional fixed fire extinguishing devices. While ensuring fire extinguishing effectiveness, it significantly reduces the amount of extinguishing agent used. Additionally, by combining the characteristics of aerosol fire extinguishing technology, it further improves the space utilization rate of the extinguishing agent.

[0016] This method calculates the optimal horizontal deflection angle and vertical pitch angle of the directional guide nozzle based on the predicted fire source location coordinates and spread direction, combined with the airflow distribution in the diesel locomotive protection area. It also incorporates computational fluid dynamics simulation corrections and fire spread prediction corrections to ensure the nozzle spray angle matches the actual fire source location and fire spread trend. This overcomes the limitations of traditional fire extinguishing devices, which only provide uniform spraying and cannot achieve precise directional spraying. Furthermore, by using a proportional-integral-derivative control algorithm to drive an electric adjustment mechanism, the nozzle is precisely adjusted to the target angle, enabling the aerosol extinguishing agent to be directly sprayed into the core area of ​​the fire source and along the critical path of fire spread. This significantly improves the contact efficiency between the extinguishing agent and the fire source, effectively shortens the extinguishing time, and significantly enhances the fire suppression effect. It can quickly control the spread of open flames and prevent the fire from expanding and causing greater equipment damage.

[0017] This method optimizes the generation of a complete fire extinguishing command set, including extinguishing dosage, spray angle, and activation time, achieving precise coordination of extinguishing timing, spray angle, and dosage release. During the fire extinguishing execution phase, the nozzle angle is adjusted before the extinguishing agent release is triggered. Simultaneously, the actual released dosage and nozzle angle feedback are monitored in real time to generate an execution deviation correction signal, ensuring a high degree of consistency between the fire extinguishing action and the command parameters. Combined with the automatic activation mode of the fixed thermal aerosol fire extinguishing device, it can promptly initiate fire extinguishing after fire identification or pre-activate based on fire spread prediction. This solves the problems of slow response and delayed operation in traditional manual fire extinguishing methods, significantly improving the automation level and overall reliability of the fire extinguishing system, and is suitable for the complex operating conditions of diesel locomotives, including high temperature, vibration, and confined spaces. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic flowchart illustrating a method according to an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, features, and advantages of this invention more apparent and understandable, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the specific embodiments. Obviously, the embodiments described below are only some embodiments of this invention, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention.

[0022] Figure 1 This is a schematic flowchart illustrating a fire prevention and extinguishing method for internal combustion locomotives provided by the present invention. The order of steps in this flowchart can be changed, and some steps can be omitted, depending on different requirements.

[0023] like Figure 1 As shown, the method includes: S1. Real-time collection of multi-source environmental signals in the protection area of ​​the diesel locomotive using point-type ultraviolet flame detectors, heat detectors, and smoke detectors; S2. Using a deep learning algorithm based on convolutional neural networks and attention mechanisms, feature fusion is performed on the multi-source environmental signals collected in step S1, and the fire identification result is output. S3. Based on the identification results of step S2, use the pre-trained cellular automata model to predict the spread trend of the fire within the protection zone of the internal combustion locomotive. S4. Based on the fire identification results of step S2 and the fire spread trend prediction results of step S3, optimize and generate fire extinguishing instructions that include extinguishing dosage and spray angle. S5. Based on the fire extinguishing command generated in step S4, control the dosage release and the spray angle of the fixed thermal aerosol fire extinguishing device installed inside the diesel locomotive to execute the fire extinguishing action.

[0024] To facilitate understanding of the present invention, the following description further illustrates the fire prevention and extinguishing method for diesel locomotives provided by the present invention, based on the principle of the fire prevention and extinguishing method for diesel locomotives and in conjunction with the fire prevention and extinguishing process for diesel locomotives in the embodiments.

[0025] Specifically, step S1 includes the following sub-steps: S11. Ultraviolet radiation signals with wavelengths ranging from 185nm to 260nm are collected using a UV nanoscale grating detector of a point-type ultraviolet flame detector. Shielding ambient stray light with wavelengths greater than 280nm, sampling frequency satisfy: ; In the formula: The sampling frequency of the ultraviolet signal is expressed in Hz. The highest frequency component of the ultraviolet signal is represented by a value of 5kHz. S12. Collect the ambient temperature signal of the power room through a temperature sensor. sampling frequency satisfy: ; In the formula: The thermal time constant has a value no greater than 1 second. S13. Collect smoke concentration signal through smoke detector. sampling frequency satisfy: ; In the formula: The velocity is the airflow speed, measured in m / s. The sampling pipe cross-sectional area is given in units of... ; The volume of the sampling cavity is expressed in units of 1000 ppm. ; S14. Timestamp the three types of signals collected in steps S11, S12, and S13. To ensure time synchronization error of multi-source signals satisfy: ; In the formula: These are the timestamps for ultraviolet, temperature, and smoke signals, respectively.

[0026] Secondly, step S2 includes: S21. The multi-source environmental signals collected in step S1 are preprocessed to obtain a standardized signal sequence. The preprocessing includes normalization. The mean and standard deviation of each detector signal are calculated by using a sliding window to eliminate dimensional differences and obtain a standardized signal sequence. S22. Input the standardized signal sequence obtained in step S21 into a one-dimensional convolutional neural network, and extract the time-frequency features of each detector signal through multi-layer convolution operations to generate ultraviolet feature matrix, temperature feature matrix and smoke feature matrix. S23. Input the three feature matrices generated in step S22 into the multi-head attention mechanism module, obtain the attention weight by calculating the similarity between the query matrix and the key matrix, perform weighted fusion on the value matrix, and output the fused feature vector. S24. Input the fused feature vector output in step S23 into the fully connected layer, calculate the fire identification probability through the Sigmoid activation function, and output the fire identification result when the fire identification probability exceeds the preset threshold.

[0027] Specifically, the normalization formula for step S21 is: ; In the formula: Let be the normalized signal of the i-th type detector at time t; These represent the ultraviolet, temperature, and smoke signals collected by the point-type ultraviolet flame detector, the temperature detector, and the smoke detector, respectively. Let be the original signal from the i-th type of detector at time t; Let be the mean value of the sliding window of the signal from the i-th type of detector; Let be the standard deviation of the sliding window for the signal from the i-th type of detector; To prevent division by zero of extremely small constants.

[0028] The convolution operation formula in step S22 is: ; In the formula: Let M be the output feature of the i-th type signal at time t in the l-th convolutional layer; ReLU is the linear rectified activation function; M is the convolutional kernel length; The weights of the convolution kernel at position m in layer l; For the l 1st floor at t The output at time m, when l=1 ; This is the bias term for the l-th layer.

[0029] The multi-head attention mechanism fusion formula in step S23 is as follows: ; Among them, the j-th attention head The calculation formula is: ; In the formula: The fused feature vector; For layer normalization operation; This is the feature concatenation operation; H is the number of attention heads; This is the output of the j-th attention head; To output the projection matrix; The concatenated feature matrix ,in These are the ultraviolet feature matrix, temperature feature matrix, and smoke feature matrix generated in step S22, respectively. These are the weight parameters for the query matrix, key matrix, and value matrix of the j-th head, respectively. is the dimension of the key vector.

[0030] The formula for calculating the fire identification probability in step S24 is as follows: ; In the formula: For fire identification probability; Use the Sigmoid activation function; This is the weight matrix of the fully connected layer; This is the bias term for the fully connected layer.

[0031] Step S2 also includes the following sub-steps: S25. Construct a deep learning model based on convolutional neural networks and attention mechanisms. Set the input layer to receive multi-source environmental signals collected in step S1, and configure three parallel convolutional branches to process ultraviolet signals, temperature signals and smoke signals respectively. S26. Stack three one-dimensional convolutional layers in each convolutional branch, followed by a batch normalization layer and a max pooling layer after each layer to extract the multi-scale time-frequency features of each detector signal and generate a feature map. S27. Input the three feature maps generated in step S26 into the feature concatenation layer for channel fusion, and then input them into a multi-head self-attention layer containing four parallel attention heads. Calculate the importance weights of each position by scaling the dot product attention and adaptively weight the features. S28. Input the weighted features from step S27 into the feedforward neural network. After mapping through two fully connected layers, generate the fire identification probability through the Sigmoid output layer to complete the forward inference process of the deep learning model.

[0032] The batch normalization formula in step S26 is as follows: ; In the formula: Features after batch normalization; The features are those obtained after the convolution operation in step S22; This represents the mean of the features in the current batch; The variance of the features in the current batch; It is the numerical stability constant; For learnable scaling parameters; These are learnable translation parameters.

[0033] The formula for calculating the scaling dot product attention in step S27 is: ; In the formula: Q is the query matrix, which is obtained by linear transformation of the feature map generated in step S26; K is the key matrix, which is obtained by linear transformation of the feature map generated in step S26; V is the value matrix, which is obtained by linear transformation of the feature map generated in step S26. The dimension of the key vector is used to scale the dot product result.

[0034] The formula for calculating the feedforward neural network in step S28 is: ; In the formula: FFN(x) is the output of the feedforward neural network; These are the weight matrices for the first and second fully connected layers, respectively. , respectively, are the bias terms of the first and second fully connected layers; x is the feature after attention weighting in step S27.

[0035] Step S27 includes the following sub-steps: S271. The three feature maps generated in step S26 are concatenated along the channel dimension to form a combined feature tensor. The query tensor, key tensor and value tensor are generated by linear projection transformation. S272. Divide the query tensor, key tensor and value tensor generated in step S271 into four sub-tensors. Each sub-tensor corresponds to an attention head, and calculate the scaling dot product attention score independently. S273. Mask the original attention score calculated for each attention head to block out information about future time steps, and normalize it using the Softmax function to obtain the attention weight matrix. S274. Perform matrix multiplication on the attention weight matrix and value tensor obtained in step S273, and concatenate and linearly project the outputs of each head to generate the final weighted fusion feature.

[0036] The segmentation formula in step S272 is as follows: ; In the formula: Let Q, K, and V be the query sub-tensor, key tensor, and value sub-tensor of the j-th attention head, respectively; Q, K, and V are the query tensor, key tensor, and value tensor generated in step S271, respectively. These are the projection weight matrices for the j-th head.

[0037] The formula for calculating the mask Softmax in step S273 is as follows: ; In the formula: Let M be the attention weight matrix for the j-th attention head; M is the mask matrix, which takes negative infinity for future time step positions and zero for current and past time step positions. Let be the dimension of the key tensor. ,in H represents the model dimension, and H represents the number of attention heads.

[0038] The splicing projection formula in step S274 is: ; The output of each head is: ; In the formula: This is the output of the multi-head attention mechanism; This is a channel-level splicing operation; This is the output of the j-th attention head; To output the linear projection weight matrix.

[0039] Next, step S3 includes: S31. Determine the initial parameters of the fire source based on the fire identification results output in step S2, including the fire source location coordinates, initial heat release rate and initial fire source radius, and initialize the cellular state matrix of the cellular automata model. S32. Discretize the protection area of ​​the internal combustion locomotive into a three-dimensional cellular mesh. Each cell records the temperature state, combustion state and fuel density state, and establishes a mapping relationship between the cell state and the initial parameters of the fire source in step S31. S33. Based on the initial heat release rate and environmental parameters of the protected area determined in step S31, calculate the fire spread velocity vector and determine the transition probability of the cell state from the safe state to the heated state. S34. Update the cell state matrix based on the transition probability calculated in step S33, iteratively calculate the cell temperature change and combustion state expansion at each time step, and predict the fire spread range at multiple future time steps. S35. Based on the cell state matrix updated in step S34, extract the boundary envelope of the open flame cell set, calculate the radius and volume of the hazardous area, and output the fire spread trend prediction results.

[0040] Furthermore, the formula for calculating the initial heat release rate of the fire source in step S31 is: ; In the formula: The initial heat release rate of the fire source is expressed in kW. The preset maximum heat release rate has a value ranging from 1000 to 5000 kW; The fire identification probability is output in step S2. The fire source growth coefficient ranges from 0.3 to 0.8. The formula for calculating the fire spread rate in step S33 is: ; In the formula: This is the fire spread velocity vector, in m / s; The baseline propagation velocity ranges from 0.01 to 0.05 m / s; The radiative heat feedback coefficient has a value ranging from 0.3 to 0.6. The heat release rate of the fire source at time t is determined by step S31. Calculated based on the fire source growth model; The thermal conductivity of air is taken as 0.026 W / (m·K); The ambient temperature is collected by the temperature sensor in step S1, and the unit is °C. The area of ​​the fire source is calculated from the initial fire source radius in step S31, and the unit is m². This represents the ambient airflow velocity vector, in m / s. The unit vector is the direction of propagation; The formula for calculating the cell state transition probability in step S34 is: ; In the formula: The cell state transition probability; Δt represents the cell state at time t, where 0 indicates a safe state and 1 indicates a heated state; Δt is the time step in seconds; and Δx is the cell size in meters. The formula for calculating the radius of the hazardous area in step S35 is: ; In the formula: For the predicted time The radius of the danger zone, in meters; The initial fire source radius determined in step S31, in meters; For the time domain of prediction, the value range is from 10s to 30s; For integration time; is the turbulent diffusion coefficient, with a value ranging from 0.1 to 0.3 m / s.

[0041] Further, step S34 includes: S341. Construct a state transition rule base for the cellular automaton model, define four basic states of the cell including safe state, heated state, open flame state and extinguished state, and establish a mapping relationship between the state transition conditions and the transition probabilities calculated in step S33. S342. Initialize the cellular space topology, divide the protected area into a regular cubic cell array, assign fuel property parameters and thermodynamic property parameters to the cells, and construct the adjacency relationship matrix between cells. S343. Based on the state transition rule base defined in step S341 and the cellular space topology constructed in step S342, perform iterative evolution calculations. At each time step, update the state variables of all cells according to the transition probability calculated in step S33, and record the spatiotemporal evolution process of the cell temperature field and the combustion front. S344. Post-process the spatiotemporal sequence of the cellular state obtained by the iterative evolution in step S343, extract the key feature parameters of fire spread, including spread speed, spread direction and dangerous area boundary, and generate the fire spread trend prediction result required in step S4.

[0042] Furthermore, the state transition rule definition formula for step S341 is as follows: ; In the formula: Let be the cell state at time t+Δt; This is the state transition function; Let t be the cell state at time t; Let t be the cell temperature at time t, in °C. This is the cellular fuel density, expressed in kg / m³. The oxygen volume fraction in the unit cell; The unit airflow velocity vector is determined by the environmental parameters of the protected area in step S33, and the unit is m / s; The transition probability of the cell state transitioning from the safe state to the heated state is calculated in step S33. The formula for calculating the cell adjacency relationship in step S342 is as follows: ; In the formula: as the central cell The set of adjacent cells; The index coordinates of the cell in the three-dimensional cell array; The cell temperature update formula in step S343 is: ; In the formula: Let be the cell temperature at time t+Δt; The heat received by a cell from neighboring cells, including thermal conduction and thermal radiation components, is expressed in J. This represents the heat lost by the cell to the environment, expressed in J. The specific heat capacity of air at constant pressure is taken as 1005 J / (kg·K); The density of air is taken as 1.2 kg / m³. The cell volume is calculated from the cell size in step S342, and the unit is m³.

[0043] Further, step S343 includes: S3431. At the start of the current time step, traverse all cells, check the transition conditions of each cell according to the state transition rule base defined in step S341, and calculate the heat radiation flux of the adjacent open flame cell to the cell. S3432. Based on the heat radiation flux calculated in step S3431 and the current cell temperature, determine whether the state transition conditions are met. If the cell temperature exceeds the ignition temperature and the oxygen concentration is higher than the critical value, then update the cell state from the heated state to the open flame state. S3433. Perform thermal balance calculations on all cells updated in step S3432, update the temperature field based on heat conduction and heat convection exchange between cells, and advance the time step to the next iteration cycle. S3434. Repeat steps S3431 to S3433 until the preset prediction time domain is reached, and output complete cell state spatiotemporal evolution data to step S344.

[0044] Furthermore, the formula for calculating the thermal radiation flux in step S3431 is as follows: ; In the formula: This represents the thermal radiation flux received by the current cell, expressed in kW / m². The set of adjacent open flame cells is determined by the adjacency matrix in step S342; The heat release rate of the m-th adjacent open flame cell is calculated from the initial heat release rate determined in step S33 and the fire source growth model, and the unit is kW; This is the viewing angle coefficient, with a value ranging from 0 to 1; The center distance between the current cell and the m-th adjacent open flame cell is calculated from the cell space coordinates in step S342, and the unit is m; The state transition judgment formula for step S3432 is: ; In the formula: The updated cell state is represented by 2 for open flame, 1 for heated state, and 0 for safe state. The fuel ignition temperature is determined by the fuel property parameters allocated in step S342, and the unit is °C. The critical oxygen volume fraction ranges from 0.15 to 0.18. The cell size defined in step S342, in meters; The heat balance calculation formula for step S3433 is as follows: ; In the formula: This represents the total heat received by the cell, expressed in J. The cell surface area is calculated from the cell size in step S342, and the unit is m². The equivalent thermal conductivity is determined by the thermodynamic property parameters allocated in step S342, and the unit is W / (m·K). The temperature of the adjacent cell n is calculated iteratively from the previous time step in step S343, and the unit is ℃; The inter-cell contact area is determined by the cell size in step S342, and the unit is m². The convective heat transfer coefficient is determined by the environmental parameters of the protected area in step S33, and the unit is W / (m²·K). The temperature of the air surrounding the cell is obtained by interpolation of data collected by the temperature sensor in step S1, and the unit is °C.

[0045] Then, step S4 includes: S41. Determine the fire intensity level based on the fire identification results output in step S2, and calculate the adaptive extinguishing dose by combining the dangerous area volume in the fire spread trend prediction results in step S3. S42. Based on the fire spread trend prediction results in step S3, determine the fire extinguishing activation timing and generate a fire extinguishing sequence command containing the activation time. S43. Based on the fire source location coordinates and spread direction in the fire spread trend prediction results predicted in step S3, and combined with the airflow field distribution in the diesel locomotive protection area, calculate the optimal injection angle of the directional guide nozzle. S44. The adaptive extinguishing dosage calculated in step S41, the extinguishing timing command determined in step S42, and the optimal spray angle calculated in step S43 are encapsulated into commands to generate a complete extinguishing command set and transmit it to the fixed thermal aerosol extinguishing device.

[0046] Furthermore, the formula for calculating the adaptive extinguishing dose in step S41 is as follows: ; In the formula: The unit is adaptive extinguishing dosage, expressed in kg; The baseline extinguishing dose is 3.0 kg; γ is the fire intensity correction factor, ranging from 0.2 to 0.5. The fire intensity level is the fire identification probability output from step S2. The values ​​obtained through mapping range from 1 to 5; It is a non-linear exponent, with a value range of 0.8 to 1.2; The predicted time for step S3 The volume of the hazardous area, in m³; For reference, the volume of the protected area is taken as 10m³; This is the leakage compensation coefficient; For prediction and correction coefficients; The formula for calculating the fire extinguishing activation time in step S42 is: ; In the formula: This is the moment when fire suppression is initiated; The current moment; Delay time for automatic mode; The automatic start threshold is set to 0.85. This is a manually triggered time. To predict the start-up time, when the radius of the danger zone predicted in step S3... Exceeding the threshold Start in advance; This is the threshold for the danger radius, with a value ranging from 2 to 5 meters. The formula for calculating the optimal injection angle in step S43 is: ; ; In the formula: The optimal horizontal deflection angle is expressed in rad. The optimal vertical pitch angle is expressed in rad. The coordinates of the fire source location predicted in step S3 are in meters. The coordinates of the directional guide nozzle installation location are in meters. This is the horizontal angle correction amount based on computational fluid dynamics simulation; This is the vertical angle correction amount based on computational fluid dynamics simulation; This is the correction amount for the predicted level based on the spread prediction in step S3; This is the predicted vertical correction amount based on the spread prediction in step S3.

[0047] In addition, step S4 includes the following sub-steps: S45. Construct a multi-objective optimization decision model, establish a decision variable space including fire extinguishing dosage, spray angle and activation timing, and set a multi-objective optimization function for fire extinguishing success rate, response time and agent utilization rate; S46. Input the fire identification results from step S2 and the fire spread trend prediction results from step S3 into the multi-objective optimization decision model as constraints and initial boundary conditions, and use the genetic algorithm to solve the Pareto optimal solution set. S47. Based on the Pareto optimal solution set obtained in step S46, and combined with the current operating parameters of the diesel locomotive and the physical constraints of the fixed thermal aerosol fire extinguishing device, select the optimal compromise solution through fuzzy decision-making. S48. Decode the optimal compromise selected in step S47 into executable fire extinguishing control parameters, including specific aerosol release amount, directional nozzle angle adjustment amount, and activation delay time, to form a fire extinguishing command.

[0048] The multi-objective optimization function in step S45 is: ; ; In the formula: The objective function vector; The decision variable vector includes aerosol dosage, horizontal injection angle, vertical injection angle, and start-up delay; With the lowest fire extinguishing success rate, The fire extinguishing success rate predicted by the cellular automata model in step S3; The system response time is expressed in seconds (s). To reduce the rate of drug waste, To predict wasted dosage, This represents the total released dose.

[0049] The fitness function of the genetic algorithm in step S46 is: ; In the formula: Let be the weight coefficient of the k-th objective, and ; These are the maximum and minimum values ​​of the k-th target, respectively; Let be the penalty coefficient for the j-th constraint; The j-th constraint is determined by the fire identification result of step S2 and the spread prediction result of step S3; J is the total number of constraints.

[0050] The fuzzy decision selection formula in step S47 is: ; In the formula: To solve The optimal membership degree; Let be the fuzzy membership degree of the k-th objective function; Let be the fuzzy membership degree of the satisfaction of the j-th constraint.

[0051] Step S46 includes the following sub-steps: S461. Initialize the population of the genetic algorithm. Determine the initial range of decision variables based on the fire identification results of step S2 and the fire spread trend prediction results of step S3, and randomly generate an initial population containing multiple individuals. S462. Evaluate the fitness of each individual in the population generated in step S461, call the cellular automata model in step S3 to simulate the fire extinguishing effect corresponding to the individual, and calculate the multi-objective optimization function value and the degree of constraint violation. S463. Based on the fitness values ​​calculated in step S462, perform non-dominated sorting, select Pareto front individuals to perform crossover and mutation operations, generate a new generation of population, and iterate until the preset maximum number of generations is reached. S464. Extract the non-dominated solution set from the Pareto front of the final population in step S463, and output it as the Pareto optimal solution set to step S47.

[0052] The population initialization formula in step S461 is as follows: ; In the formula: This is the i-th initial individual; A uniformly distributed random number between 0 and 1; These are the disturbance coefficients for each decision variable, ranging from 0.1 to 0.3. The initial injection angle is estimated from the fire source location coordinates in step S3.

[0053] The cellular automaton simulation call formula in step S462 is: ; In the formula: This is the cell state matrix at the end of the simulation; The cellular automaton model pre-trained in step S3; The initial cell state is determined by the fire spread trend prediction result in step S3; The fire extinguishing control parameters corresponding to the individual to be evaluated; The simulation time step is expressed in seconds. The total duration is simulated in seconds.

[0054] The formula for the crossover operation in step S463 is: ; ; In the formula: Offspring individuals generated through crossover; The parent individuals are randomly selected from the Pareto front; This is the cross coefficient, with a value ranging from 0.3 to 0.7.

[0055] Finally, step S5 includes: S51. Analyze the fire extinguishing command generated in step S4, extract the adaptive fire extinguishing dosage, optimal spray angle and start-up time control parameters, and verify the completeness and effectiveness of the command; S52. Based on the optimal injection angle analyzed in step S51, the electric adjustment mechanism is driven by the proportional-integral-derivative control algorithm to adjust the horizontal deflection angle and vertical pitch angle of the directional guide nozzle to the target position. S53. After the adjustment in step S52 is completed, the fixed thermal aerosol fire extinguishing device is triggered according to the start time analyzed in step S51, and the graded dose release is performed. The first stage releases large-diameter aerosol particles to quickly suppress open flames, and the second stage releases small-diameter aerosol particles to maintain the suspension concentration. S54. Monitor the actual release dose and nozzle angle feedback during the execution of step S53, compare them in real time with the target parameters analyzed in step S51, generate an execution deviation correction signal, and ensure that the fire extinguishing action is executed accurately.

[0056] Specifically, the proportional-integral-derivative control algorithm formula in step S52 is as follows: ; ; In the formula: These are the control output quantities in the horizontal and vertical directions at time t, respectively; These are the proportional coefficient, integral coefficient, and differential coefficient, with values ​​ranging from 5 to 20, 0.1 to 1, and 0.5 to 2, respectively. For horizontal angle deviation, The optimal horizontal spray angle generated in step S4 Let t be the actual horizontal angle of the directional nozzle. This is the vertical angle deviation. The optimal vertical injection angle generated in step S4. The actual vertical angle of the directional guide nozzle at time t; This is the time variable for integration.

[0057] The formula for graded dose release control in step S53 is: ; ; In the formula: These are the cumulative release doses for the first and second stages, respectively, in kg. These are the release rates of the first and second stages, respectively, in kg / s; These represent the release durations of the first and second stages, respectively, in seconds. This represents the percentage of the first-level release, ranging from 0.6 to 0.8. The adaptive extinguishing dose generated in step S4 is expressed in kg.

[0058] The formula for correcting execution deviation in step S54 is: ; ; In the formula: This is a dose correction measure, in kg. This is the angle correction amount, in rad; These are the dose ratio correction factor, the dose integral correction factor, and the angle correction factor, respectively. These are the actual released dose and the target dose analyzed in step S51, respectively. These are the actual injection angle and the target angle analyzed in step S51, respectively.

[0059] Step S5 includes the following sub-steps: S55. Establish a digital twin model of a fixed thermal aerosol fire extinguishing device, integrate the agent combustion dynamics model, the nozzle airflow field model and the electromechanical coupling model of the electric adjustment mechanism, and construct a virtual simulation environment; S56. Input the fire extinguishing command generated in step S4 into the digital twin model established in step S55 to perform fire extinguishing action pre-simulation and predict the fire extinguishing effect and equipment status response under different execution strategies. S57. Based on the simulation results of step S56, optimize and adjust the execution parameters of the fire extinguishing command, including the correction of the release rate curve, the fine adjustment of the spray angle and the compensation of the start time, and generate the optimized execution strategy. S58. Following the optimized execution strategy in step S57, drive the fixed thermal aerosol fire extinguishing device and the directional flow nozzle to perform actual fire extinguishing actions, and simultaneously collect execution feedback data to update the state of the digital twin model.

[0060] The combustion kinetics model formula for the agent in step S55 is as follows: ; In the formula: The mass of aerosols generated is expressed in kg. The burning area of ​​the agent is expressed in m². Density of the aerosol generator, in kg / m³; The combustion rate is denoted as , and the combustion chamber pressure is denoted as . and temperature The function is expressed in m / s.

[0061] The objective function for the pre-simulation in step S56 is: ; In the formula: For the evaluation indicators of pre-simulation; The simulated total release dose is expressed in kg. The adaptive extinguishing dose generated in step S4, in kg; The simulated peak concentration is expressed in kg / m³. The target extinguishing concentration is expressed in kg / m³. Coverage time, in seconds; These are the weighting coefficients.

[0062] The optimization adjustment formula for step S57 is: ; In the formula: η is the execution parameter vector for the k-th iteration, including the release rate and injection angle; η is the learning rate, ranging from 0.01 to 0.1. This is to evaluate the gradient of the performance index with respect to the execution parameters.

[0063] In addition, step S56 includes the following sub-steps: S561. Initialize the state of the digital twin model according to the fire extinguishing command generated in step S4, set the initial mass of the agent, the initial angle of the nozzle and the environmental boundary conditions, and establish the initial state of the simulation. S562. Perform iterative simulation calculations, solve the agent combustion kinetics equation, aerosol diffusion and convection equation and electric regulating mechanism motion equation at each time step, and update the system state variables. S563. During the iteration process of step S562, monitor key performance indicators, including the concentration distribution in the protected area, temperature change gradient, and degree of open flame suppression, and determine whether the fire extinguishing termination conditions are met. S564. Output the complete simulation timing data of steps S562 and S563, calculate the fire extinguishing success rate, response time, and agent utilization rate, and generate a pre-simulation result report to be transmitted to step S57.

[0064] The aerosol diffusion and convection equation in step S562 is as follows: ; In the formula: C is the aerosol concentration, in kg / m³; t is the time, in seconds; The airflow velocity vector is determined by the nozzle airflow field model in step S55, and its unit is m / s; The turbulent diffusion coefficient is expressed in m² / s. For gradient operators; The aerosol source term is determined by the output of the agent combustion kinetics model in step S55, and the unit is kg / (m³·s); This represents the aerosol sedimentation accumulation, expressed in kg / (m³·s).

[0065] The fire extinguishing termination condition in step S563 is: ; In the formula: The highest temperature in the protected area at time t is determined by the temperature field obtained in step S562, and the unit is °C. The safe temperature threshold ranges from 60 to 80℃. The area of ​​the open flame region at time t is obtained from the statistical analysis of the cellular combustion state solved in step S562, and the unit is m². The average aerosol concentration in the protected area at time t is obtained by averaging the concentration field obtained in step S562, with units of kg / m³. The minimum effective extinguishing concentration is defined as 0.05 to 0.1 kg / m³.

[0066] The formula for calculating the fire extinguishing success rate in step S564 is as follows: ; In the formula: To simulate the success rate of fire extinguishing; These are the highest temperatures at the beginning and end of the simulation, respectively, in °C. These are the open flame areas at the beginning and end of the simulation, respectively, in m². The effective concentration maintenance time is expressed in seconds. These are the weighting coefficients for temperature suppression, open flame extinguishing, and concentration maintenance, respectively, and the sum of the three is 1.

[0067] Although the present invention has been described in detail with reference to the accompanying drawings and preferred embodiments, the present invention is not limited thereto. Various equivalent modifications or substitutions can be made to the embodiments of the present invention by those skilled in the art without departing from the spirit and essence of the invention, and such modifications or substitutions should all be within the scope of the present invention. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should also be covered within the protection scope of the present invention.

Claims

1. A method for fire prevention and extinguishing in an internal combustion locomotive, characterized in that, include: S1. Real-time collection of multi-source environmental signals in the protection area of ​​the diesel locomotive using point-type ultraviolet flame detectors, heat detectors, and smoke detectors; S2. Using a deep learning algorithm based on convolutional neural networks and attention mechanisms, feature fusion is performed on the multi-source environmental signals collected in step S1, and the fire identification result is output. S3. Based on the identification results of step S2, use the pre-trained cellular automata model to predict the spread trend of the fire within the protection area of ​​the internal combustion locomotive. S4. Based on the fire identification results of step S2 and the fire spread trend prediction results of step S3, optimize and generate fire extinguishing instructions that include extinguishing dosage and spray angle. S5. Based on the fire extinguishing command generated in step S4, control the dosage release and the spray angle of the fixed thermal aerosol fire extinguishing device installed inside the diesel locomotive to execute the fire extinguishing action.

2. The fire prevention and extinguishing method for internal combustion locomotives according to claim 1, characterized in that, Step S2 includes: S21. Preprocess the multi-source environmental signals collected in step S1 to obtain a standardized signal sequence; S22. Input the standardized signal sequence obtained in step S21 into a one-dimensional convolutional neural network, and extract the time-frequency features of each detector signal through multi-layer convolution operations to generate ultraviolet feature matrix, temperature feature matrix and smoke feature matrix. S23. Input the three feature matrices generated in step S22 into the multi-head attention mechanism module, obtain the attention weight by calculating the similarity between the query matrix and the key matrix, perform weighted fusion on the value matrix, and output the fused feature vector. S24. Input the fused feature vector output in step S23 into the fully connected layer, calculate the fire identification probability through the Sigmoid activation function, and output the fire identification result when the fire identification probability exceeds the preset threshold.

3. The fire prevention and extinguishing method for internal combustion locomotives according to claim 1, characterized in that, Step S3 includes: S31. Determine the initial parameters of the fire source based on the fire identification results output in step S2, including the fire source location coordinates, initial heat release rate and initial fire source radius, and initialize the cellular state matrix of the cellular automata model. S32. Discretize the protection area of ​​the internal combustion locomotive into a three-dimensional cellular mesh. Each cell records the temperature state, combustion state and fuel density state, and establishes a mapping relationship between the cell state and the initial parameters of the fire source in step S31. S33. Based on the initial heat release rate and environmental parameters of the protected area determined in step S31, calculate the fire spread velocity vector and determine the transition probability of the cell state from the safe state to the heated state. S34. Update the cell state matrix based on the transition probability calculated in step S33, iteratively calculate the cell temperature change and combustion state expansion at each time step, and predict the fire spread range at multiple future time steps. S35. Based on the cell state matrix updated in step S34, extract the boundary envelope of the open flame cell set, calculate the radius and volume of the hazardous area, and output the fire spread trend prediction results.

4. The fire prevention and extinguishing method for internal combustion locomotives according to claim 3, characterized in that, The formula for calculating the initial heat release rate of the fire source in step S31 is: ； In the formula: The initial heat release rate of the fire source; The preset maximum heat release rate; The fire identification probability is output in step S2. The fire source growth coefficient; The formula for calculating the fire spread rate in step S33 is: ； In the formula: This is the fire spread velocity vector; Based on the spread rate; The radiative heat feedback coefficient; Let t be the heat release rate of the fire source. The thermal conductivity of air; Ambient temperature; Area of ​​the fire source; This represents the ambient airflow velocity vector. The unit vector represents the direction of propagation. The formula for calculating the cell state transition probability in step S34 is: ； In the formula: The cell state transition probability; Let Δt be the cell state at time t, where 0 represents the safe state and 1 represents the heated state; Δt is the time step; and Δx is the cell size. The formula for calculating the radius of the hazardous area in step S35 is: ； In the formula: For the predicted time The radius of the danger zone; The initial fire source radius determined in step S31; For prediction in the time domain; For integration time; is the turbulent diffusion coefficient.

5. The fire prevention and extinguishing method for internal combustion locomotives according to claim 3, characterized in that, Step S34 includes: S341. Construct a state transition rule base for the cellular automaton model, define four basic states of the cell including safe state, heated state, open flame state and extinguished state, and establish a mapping relationship between the state transition conditions and the transition probabilities calculated in step S33. S342. Initialize the cellular space topology, divide the protected area into a regular cubic cell array, assign fuel property parameters and thermodynamic property parameters to the cells, and construct the adjacency relationship matrix between cells. S343. Based on the state transition rule base defined in step S341 and the cellular space topology constructed in step S342, perform iterative evolution calculations. At each time step, update the state variables of all cells according to the transition probability calculated in step S33, and record the spatiotemporal evolution process of the cell temperature field and the combustion front. S344. Post-process the spatiotemporal sequence of the cellular state obtained by the iterative evolution in step S343, extract the key feature parameters of fire spread, including spread speed, spread direction and dangerous area boundary, and generate the fire spread trend prediction result required in step S4.

6. The fire prevention and extinguishing method for internal combustion locomotives according to claim 5, characterized in that, Step S343 includes: S3431. At the start of the current time step, traverse all cells, check the transition conditions of each cell according to the state transition rule base defined in step S341, and calculate the heat radiation flux of the adjacent open flame cell to the cell. S3432. Based on the heat radiation flux calculated in step S3431 and the current cell temperature, determine whether the state transition conditions are met. If the cell temperature exceeds the ignition temperature and the oxygen concentration is higher than the critical value, then update the cell state from the heated state to the open flame state. S3433. Perform thermal balance calculations on all cells updated in step S3432, update the temperature field based on heat conduction and heat convection exchange between cells, and advance the time step to the next iteration cycle. S3434. Repeat steps S3431 to S3433 until the preset prediction time domain is reached, and output complete cell state spatiotemporal evolution data to step S344.

7. The fire prevention and extinguishing method for internal combustion locomotives according to claim 6, characterized in that, The formula for calculating the thermal radiation flux in step S3431 is: ； In the formula: This represents the thermal radiation flux received by the current cell. The set of adjacent open flame cells is determined by the adjacency matrix in step S342; Let be the heat release rate of the m-th adjacent open flame cell; For the perspective coefficient; The distance between the center of the current cell and the m-th adjacent open flame cell; The state transition judgment formula for step S3432 is: ； In the formula: The updated cell state is represented by 2 for open flame, 1 for heated state, and 0 for safe state. This refers to the fuel ignition temperature. The critical oxygen volume fraction; The cell size defined in step S342; The heat balance calculation formula for step S3433 is as follows: ； In the formula: This represents the total heat received by the cell. The surface area of ​​the cell; Equivalent thermal conductivity; The temperature of the adjacent cell n; This represents the contact area between cells; The convective heat transfer coefficient; The temperature of the air surrounding the cell.

8. The fire prevention and extinguishing method for internal combustion locomotives according to claim 3, characterized in that, Step S4 includes: S41. Determine the fire intensity level based on the fire identification results output in step S2, and calculate the adaptive extinguishing dose by combining the dangerous area volume in the fire spread trend prediction results in step S3. S42. Based on the fire spread trend prediction results in step S3, determine the fire extinguishing activation timing and generate a fire extinguishing sequence command containing the activation time. S43. Based on the fire source location coordinates and spread direction in the fire spread trend prediction results predicted in step S3, and combined with the airflow field distribution in the diesel locomotive protection area, calculate the optimal injection angle of the directional guide nozzle. S44. The adaptive extinguishing dosage calculated in step S41, the extinguishing timing command determined in step S42, and the optimal spray angle calculated in step S43 are encapsulated into commands to generate a complete extinguishing command set and transmit it to the fixed thermal aerosol extinguishing device.

9. The fire prevention and extinguishing method for internal combustion locomotives according to claim 8, characterized in that, The formula for calculating the adaptive extinguishing dose in step S41 is: ； In the formula: For adaptive extinguishing dosage; The baseline extinguishing dose is γ; γ is the fire intensity correction factor. The fire intensity level is the fire identification probability output from step S2. The value obtained through mapping ranges from 1 to 5; It is a non-linear exponent; The predicted time for step S3 The volume of the danger zone; For reference, the volume of the protected area is used. This is the leakage compensation coefficient; For prediction and correction coefficients; The formula for calculating the fire extinguishing activation time in step S42 is: ； In the formula: This is the moment when fire suppression is initiated; The current moment; Delay time for automatic mode; The threshold for automatic startup; This is a manually triggered time. To predict the start time, when the radius of the danger zone predicted in step S3... Exceeding the threshold Start in advance; This is the threshold for the danger radius; The formula for calculating the optimal injection angle in step S43 is: ； ； In the formula: The optimal horizontal deflection angle; To achieve the optimal vertical pitch angle; The coordinates of the fire source location predicted in step S3; To determine the coordinates of the guide nozzle installation position; This is the horizontal angle correction amount based on computational fluid dynamics simulation; This is the vertical angle correction amount based on computational fluid dynamics simulation; This is the correction amount for the predicted level based on the spread prediction in step S3; This is the predicted vertical correction amount based on the spread prediction in step S3.

10. The fire prevention and extinguishing method for internal combustion locomotives according to claim 8, characterized in that, Step S5 includes: S51. Analyze the fire extinguishing command generated in step S4, extract the adaptive fire extinguishing dosage, optimal spray angle and start-up time control parameters, and verify the completeness and effectiveness of the command; S52. Based on the optimal injection angle analyzed in step S51, the electric adjustment mechanism is driven by the proportional-integral-derivative control algorithm to adjust the horizontal deflection angle and vertical pitch angle of the directional guide nozzle to the target position. S53. After the adjustment in step S52 is completed, the fixed thermal aerosol fire extinguishing device is triggered according to the start time analyzed in step S51, and the graded dose release is performed. The first stage releases large-diameter aerosol particles to quickly suppress open flames, and the second stage releases small-diameter aerosol particles to maintain the suspension concentration. S54. Monitor the actual release dose and nozzle angle feedback during the execution of step S53, compare them in real time with the target parameters analyzed in step S51, generate an execution deviation correction signal, and ensure that the fire extinguishing action is executed accurately.

Images