Intelligent monitoring operation method and system for tunnel highway fire fighting

Abstract

The application discloses an intelligent monitoring operation method and system for tunnel highway fire fighting, relates to the technical field of tunnel fire safety, and comprises the following steps: inputting a vibration feature vector into a pre-trained fire classification model to obtain a fire event, and calculating a fire source position through time domain reflection positioning; constructing a deep reinforcement learning model based on the fire source position and tunnel state parameters, generating a fire-fighting robot path and a water pump start-stop instruction; controlling the fire-fighting robot to move according to the fire-fighting robot path and the water pump start-stop instruction and adjust water pump injection parameters, collecting fire spread data in real time to generate a feedback signal; updating the reinforcement learning model parameters according to the feedback signal, dynamically optimizing the fire-fighting robot path and the water pump start-stop instruction, and re-executing; and extracting accurate vibration feature vectors by using wavelet denoising processing, so that high-sensitivity monitoring and early warning of fire hazards are realized, and the intelligent level and emergency response capability of tunnel highway fire prevention and control are significantly improved.

Description

Technical Field

[0001] This invention relates to the field of tunnel fire safety technology, and in particular to an intelligent monitoring and operation method and system for fire prevention in tunnel highway fires. Background Technology

[0002] In recent years, with the continuous advancement of fiber optic sensing technology, artificial intelligence algorithms, and automated control systems, tunnel fire monitoring systems have gradually developed towards intelligent and multimodal sensing. Existing technologies have employed distributed fiber optics to collect temperature or vibration signals, combined with image recognition and smoke detection to achieve early fire warnings. Pre-set control logic drives fire-fighting equipment for emergency response, improving the accuracy and response speed of tunnel fire monitoring to some extent. However, the narrow, elongated space and limited ventilation within tunnels allow fires to spread rapidly, posing significant challenges to personnel evacuation and firefighting rescue efforts.

[0003] Monitoring methods for tunnel fires mainly focus on monitoring parameters such as temperature and smoke concentration. These methods often fail to provide sufficient early warning information, especially in the early stages of a fire, and are prone to false alarms or missed alarms, resulting in low efficiency in the dispatch of fire-fighting resources. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides an intelligent monitoring and operation method for fire prevention in tunnel highways to solve the problem of low accuracy in fire source identification.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides an intelligent monitoring and operation method for fire fighting in tunnel highways, comprising: collecting vibration signals through distributed optical fibers on the tunnel sidewalls and extracting vibration features by performing wavelet denoising processing to obtain vibration feature vectors; inputting the vibration feature vectors into a pre-trained fire classification model to obtain fire events, and calculating the fire source location through time-domain reflection positioning; constructing a deep reinforcement learning model based on the fire source location and tunnel state parameters to generate fire robot paths and water pump start / stop commands; controlling the fire robot to move according to the fire robot paths and water pump start / stop commands and adjusting water pump spray parameters, and collecting fire spread data in real time to generate feedback signals; updating the reinforcement learning model parameters according to the feedback signals, dynamically optimizing the fire robot paths and water pump start / stop commands and re-executing them; setting a fire safety threshold according to the tunnel cross-sectional area, and assessing the fire status and detecting the fire extinguishing status through the fire safety threshold.

[0007] As a preferred embodiment of the intelligent monitoring and operation method for tunnel highway fire prevention described in this invention, the method involves: collecting vibration signals through distributed optical fibers on the tunnel sidewalls, and extracting vibration features through wavelet noise reduction processing to obtain a vibration feature vector. The specific steps are as follows. Vibration signals are generated by collecting data on the tunnel's structural health, external influences, and traffic load through distributed optical fibers along the tunnel sidewalls. Noise and interference signals in the vibration signal are removed by filtering, and the vibration signal is analyzed by wavelet transform to decompose the vibration signal into different frequency components. The vibration signal is then reconstructed by inverse wavelet transform to obtain the noise-reduced vibration signal. The vibration signal after noise reduction is analyzed by wavelet transform, and features in the time domain, frequency domain, and based on wavelet energy distribution are extracted. The extracted features are combined into a feature vector to obtain the vibration feature vector.

[0008] As a preferred embodiment of the intelligent monitoring and operation method for fire prevention in tunnel highways described in this invention, the following steps are taken: Vibration feature vectors are input into a pre-trained fire classification model to obtain fire events, and the location of the fire source is calculated through time-domain reflectometry. The vibration feature vector is input into a pre-trained fire classification model for analysis to obtain the fire distribution probability and output the category label. Based on the category label, the time-domain reflection positioning function of the distributed optical fiber is triggered. By sending a pulse signal to the optical fiber and receiving the reflected signal, the specific location of the fire source on the optical fiber is calculated based on the signal propagation time and reflection characteristics. Combined with the deployment path of the optical fiber on the tunnel sidewall, the location of the fire source is mapped to the actual spatial coordinates of the tunnel.

[0009] As a preferred embodiment of the intelligent monitoring and operation method for fire fighting in tunnel highways described in this invention, the following steps are taken: A deep reinforcement learning model is constructed based on the fire source location and tunnel state parameters to generate fire robot paths and water pump start / stop commands. The state space is composed of the fire source location, tunnel status parameters and fire equipment status, and the action space is defined according to the activity range of the fire robot and the switching action of the water pump. By combining deep reinforcement learning algorithms with neural networks, a deep reinforcement learning model is constructed based on the state space and action space. By combining simulation software with the actual environment of the tunnel, a virtual environment for simulating tunnel fires is constructed. In the virtual environment, the deep reinforcement learning model is optimized through continuous interaction of state-action-reward loop. The state space is input into the optimized deep reinforcement learning model to generate the movement path of the fire-fighting robot and the start / stop control commands for the water pump.

[0010] As a preferred embodiment of the intelligent monitoring and operation method for fire fighting in tunnel highways described in this invention, the method involves controlling a fire-fighting robot to move according to its path and water pump start / stop commands, adjusting water pump spray parameters, and collecting real-time fire spread feedback signals. The specific steps are as follows: The fire-fighting robot moves from its current location to the fire source according to the planned path. When the fire-fighting robot approaches the fire source, it starts the water pump and adjusts the spray parameters according to the start and stop control commands of the water pump. Data on fire spread is collected via distributed optical fiber, and the degree of fire reduction and extinguishing efficiency during the firefighting process are recorded and analyzed to evaluate the effectiveness of the firefighting strategy. The evaluation results are then combined with the operating status of the firefighting robot and the working status of the water pump as feedback signals.

[0011] As a preferred embodiment of the intelligent monitoring and operation method for fire fighting in tunnels and highways described in this invention, the following steps are taken: updating the reinforcement learning model parameters based on feedback signals, dynamically optimizing the fire robot path and water pump start / stop commands, and re-executing them. The feedback signal is input into the deep reinforcement learning model, and the deep reinforcement learning model is learned online using the feedback signal. The parameters of the deep reinforcement learning model are adjusted by the policy gradient method. Based on the updated deep reinforcement learning model, the optimized movement path of the fire-fighting robot and the start / stop control commands of the water pump are recalculated. The fire-fighting robot readjusts its route according to the optimized fire-fighting robot path, and readjusts its spray parameters according to the optimized start / stop control commands of the water pump after approaching the fire source.

[0012] As a preferred embodiment of the intelligent monitoring and operation method for fire fighting in tunnel highways described in this invention, the following steps are taken: A fire safety threshold is obtained based on the tunnel cross-sectional area, the fire status is assessed using the fire safety threshold, and multispectral scanning is initiated to detect the fire extinguishing status. Based on the tunnel's geometry, the airflow capacity and heat diffusion capacity are calculated. The fire safety threshold is obtained by combining the tunnel parameters, fire dynamics theory, and tunnel cross-sectional area with the airflow capacity and heat diffusion capacity. Distributed optical fiber is used to monitor temperature distribution, smoke concentration, and flame coverage within the tunnel. The collected data is compared with fire safety thresholds to assess the fire status. Based on the fire situation, a multispectral scan is initiated to comprehensively scan the fire area and detect the fire suppression progress.

[0013] Secondly, the present invention provides an intelligent monitoring and operation system for fire fighting in tunnel highways, comprising a data acquisition module, a positioning module, an instruction generation module, an execution module, an optimization module, and a detection module; The system comprises the following modules: a data acquisition module, which collects vibration signals via distributed optical fibers along the tunnel sidewalls and extracts vibration features through wavelet denoising to obtain vibration feature vectors; a localization module, which inputs the vibration feature vectors into a pre-trained fire classification model to obtain fire events and calculates the fire source location using time-domain reflectometry; a command generation module, which constructs a deep reinforcement learning model based on the fire source location and tunnel state parameters to generate fire robot paths and water pump start / stop commands; an execution module, which controls the fire robot to move according to the fire robot paths and water pump start / stop commands and adjusts the water pump spray parameters, while also collecting real-time fire spread data to generate feedback signals; an optimization module, which updates the reinforcement learning model parameters based on the feedback signals, dynamically optimizes the fire robot paths and water pump start / stop commands, and re-executes them; and a detection module, which sets a fire safety threshold based on the tunnel cross-sectional area, assesses the fire status using the fire safety threshold, and detects the fire extinguishing status.

[0014] Thirdly, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the intelligent monitoring and operation method for tunnel and highway fire fighting as described in the first aspect of the present invention.

[0015] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the intelligent monitoring and operation method for tunnel and highway fire fighting as described in the first aspect of the present invention.

[0016] The beneficial effects of this invention are as follows: By collecting vibration signals through distributed optical fibers on the tunnel sidewalls and extracting precise vibration feature vectors using wavelet denoising processing, highly sensitive monitoring and early warning of fire hazards are achieved, effectively reducing the false alarm rate and improving the accuracy of fire source location. Furthermore, by utilizing time-domain reflectometry to accurately locate the fire source, crucial information is provided, significantly enhancing the intelligence level and emergency response capability of tunnel highway fire prevention and control. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of an intelligent monitoring and operation method for fire fighting in tunnels and highways.

[0019] Figure 2 This is a schematic diagram of an intelligent monitoring and operation system for fire fighting in tunnels and highways.

[0020] Figure 3 This is a flowchart for extracting vibration features using wavelet denoising.

[0021] Figure 4 A flowchart for building and optimizing deep reinforcement learning models. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0025] Reference Figures 1-4 As one embodiment of the present invention, this embodiment provides an intelligent monitoring and operation method for fire fighting in tunnel highways, comprising the following steps: S1. Vibration signals are collected through distributed optical fibers on the tunnel sidewalls, and wavelet noise reduction is performed to extract vibration features, resulting in a vibration feature vector.

[0026] Vibration signals are generated by collecting data on the tunnel's structural health, external influences, and traffic loads through distributed optical fibers along the tunnel sidewalls.

[0027] Specifically, distributed optical fibers are deployed on the tunnel sidewalls. These distributed optical fibers can monitor in real time the minute vibrations caused by structural changes, external environmental factors (such as earthquakes and ground movement), and passing vehicles. Optical time domain reflectance (OTDR) technology is used to analyze the optical signals transmitted in the optical fibers to detect changes in the optical signals inside the optical fibers caused by the tunnel's structural health, external influences, and traffic load factors, thus forming vibration signals that reflect the tunnel's condition.

[0028] Noise and interference signals in the vibration signal are removed by filtering. The vibration signal is analyzed by wavelet transform and decomposed into different frequency components. The vibration signal is reconstructed by inverse wavelet transform to obtain the noise-reduced vibration signal.

[0029] Specifically, filters (such as bandpass or lowpass filters) are applied to remove obvious noise and interference signals. Wavelet transform is used to decompose the filtered vibration signal into wavelet coefficients of different frequency bands. After wavelet decomposition, an adaptive denoising threshold is obtained based on an analysis method combining signal local singularity and noise statistical characteristics. The high-frequency detail coefficients are then processed using the adaptive denoising threshold, setting high-frequency wavelet coefficients with absolute values ​​less than the adaptive denoising threshold to zero. The processed low-frequency approximation coefficients and high-frequency detail coefficients are then reconstructed using inverse wavelet transform. Simultaneously, high-frequency wavelet coefficients greater than or equal to the adaptive denoising threshold are shrunk to preserve structural information. The adjusted wavelet coefficients are then reconstructed using inverse wavelet transform to obtain the denoised vibration signal.

[0030] It should be noted that, based on the characteristics of the vibration signal (such as signal-to-noise ratio, frequency distribution, etc.), statistical analysis methods are used to estimate the noise level. For example, the noise intensity is assessed by calculating the standard deviation of the high-frequency coefficients after wavelet decomposition. A general threshold method is selected to set the threshold based on the noise intensity.

[0031] The vibration signal after noise reduction is analyzed by wavelet transform, and features in the time domain, frequency domain, and based on wavelet energy distribution are extracted. The extracted features are combined into a feature vector to obtain the vibration feature vector.

[0032] Specifically, wavelet transform is applied for multi-level decomposition to obtain details and approximation coefficients in different frequency bands, enabling detailed analysis of signal characteristics in both the time and frequency domains. Multiple features are extracted from the decomposition levels: statistical parameters such as the mean, variance, and peak value of the vibration signal are calculated in the time domain; and features such as energy distribution and dominant frequency components in each frequency band are analyzed in the frequency domain. Based on wavelet energy distribution, the wavelet energy percentage at each decomposition level is calculated to reflect the energy distribution of the vibration signal at different scales. Multiple features extracted from the time domain, frequency domain, and wavelet energy distribution are combined into a single vibration feature vector.

[0033] S2. Input the vibration feature vector into the pre-trained fire classification model to obtain fire events, and calculate the location of the fire source through time-domain reflection localization.

[0034] The vibration feature vector is input into a pre-trained fire classification model for analysis to obtain the fire distribution probability and output the category label.

[0035] Specifically, a large amount of vibration signal data is collected and labeled, including normal conditions and different types of fires. Filtering, denoising, and feature extraction are performed to form vibration feature vectors. A suitable deep learning model architecture is selected, the input is defined as the vibration feature vector, the output is the probability distribution of fire categories, cross-entropy is set as the loss function, and optimization algorithms such as Adam or SGD are used to update the parameters. The collected fire-related vibration signal data were divided into training and testing sets. The training set was used for multiple rounds of iterative training. In each round, the error between the predicted value and the true label was calculated via forward propagation, and the weights were adjusted via backpropagation. Simultaneously, a validation set was used to monitor the performance of the pre-trained fire classification model. Regularization techniques or early stopping methods were applied as needed to prevent overfitting. After all training rounds were completed, an independent testing set was used to evaluate the overall performance of the pre-trained fire classification model. The trained fire classification model and its weights were saved. The extracted vibration feature vectors were preprocessed, and their shape was adjusted to a two-dimensional array (using Python and TensorFlow frameworks). For example, normalization or standardization was used to ensure the vibration feature vectors fell within a suitable range. Finally, the model was converted into the tensor form required by the fire classification model (such as NumPy arrays or PyTorch / TensorFlow tensors). The processed vibration feature vector is input into the trained fire classification model. The pre-trained fire classification model internally uses a multi-layer neural network (such as fully connected layers and convolutional layers) to perform nonlinear transformations and feature mapping on the vibration feature vector, calculating the probability distribution of each fire category. The Softmax function is then used to convert the raw scores output by the fire classification model into probability values, representing the likelihood that the vibration signal corresponds to each type of fire (e.g., no fire, open flame, smoldering fire). Based on the probability distribution of fire categories, the category label corresponding to the highest probability is selected as the output.

[0036] Based on the category label, the time-domain reflection positioning function of the distributed optical fiber is triggered. By sending a pulse signal to the optical fiber and receiving the reflected signal, the specific location of the fire source on the optical fiber is calculated based on the signal propagation time and reflection characteristics. Combined with the deployment path of the optical fiber on the tunnel sidewall, the location of the fire source is mapped to the actual spatial coordinates of the tunnel.

[0037] Specifically, based on the category label, when the classification result is a fire event, the time-domain reflection localization function of the distributed optical fiber is immediately triggered. Short-pulse optical signals are sent to the distributed optical fibers laid along the tunnel sidewall. These short-pulse optical signals are reflected when they encounter non-uniform points in the fiber (such as areas of temperature change caused by a fire). The reflected optical signals are then received, and the arrival time of each reflected optical signal is recorded. Based on the speed of light in optical fiber (approximately 2 / 3 the speed of light in a vacuum), the distance from the reflection point to the fiber's origin is calculated. By analyzing the time differences and intensity variations of multiple reflected signals, the coordinates of the most probable fire source on the optical fiber are determined. Combined with the specific fiber optic cable layout along the tunnel sidewalls (e.g., a detailed plan of the fiber optic cable's installation along the tunnel's top or walls), the relative positions on the optical fiber are mapped to actual tunnel spatial coordinates, thus accurately locating the fire source's actual three-dimensional position.

[0038] The expression for calculating the distance to the fire source should be explained as follows:

[0039] Where L is the distance to the fire source, v is the speed of light, and t is the round-trip time of the reflected light signal.

[0040] S3. Construct a deep reinforcement learning model based on the fire source location and tunnel state parameters to generate fire robot paths and water pump start / stop commands.

[0041] The state space is composed of the actual three-dimensional location of the fire source, tunnel state parameters, and the state of the fire-fighting equipment. The action space is defined based on the activity range of the fire-fighting robot and the switching actions of the water pump.

[0042] Specifically, temperature distribution and smoke concentration data are collected in real time within the tunnel using distributed fiber optic cables, infrared thermal imagers, laser scattering smoke sensors, and high-definition cameras. Kalman filtering and mean filtering algorithms are used to clean the raw data and eliminate noise. A GPS / inertial navigation module is mounted on the fire-fighting robot to obtain its real-time location, a battery management system to read battery power information, a liquid level sensor to monitor the remaining extinguishing agent, and a water pump control circuit to obtain its on / off status. The location coordinates are standardized according to tunnel mileage, battery power and water level are converted to percentages, and the water pump status is encoded as 0 / 1 values ​​to obtain a standardized numerical representation of the equipment information. Temperature distribution and smoke concentration data from monitoring points within the tunnel are selected and cleaned. The current location of the fire-fighting robot, battery power, remaining extinguishing agent, and the on / off status of the water pump are converted into standardized numerical representations to obtain tunnel status parameters and the status of the fire-fighting equipment. For the actual three-dimensional location and intensity of the fire source, distributed optical fiber, a heat release rate formula based on the characteristics of combustion materials and mass flow rate, and multi-source data fusion algorithms such as Kalman filtering are used to convert the coordinates, spatial distribution characteristics, and heat release parameters (HRR, temperature gradient, smoke generation rate, etc.) of the fire source in the tunnel space into precise values. Accuracy is improved through real-time calibration and normalization to adapt to a unified numerical range. These values ​​are then combined sequentially (e.g., fire source information first, then environmental parameters, and finally equipment status) into a high-dimensional vector, forming a complete state space representation. A unique identifier is assigned to each fire robot position. For each position, the robot's executable action options are listed, including basic movement commands such as forward, backward, left turn, right turn, and stop. Considering different operating modes of the water pump, operations (such as starting, stopping, and adjusting water pressure) are encoded into independent action commands and integrated with the movement commands into the same action space. By constructing a joint action set, where each action consists of a position identifier plus one or more operation commands, the optimal action strategy can be selected based on the current state.

[0043] By combining deep reinforcement learning algorithms with neural networks, a deep reinforcement learning model is constructed based on the state space and action space.

[0044] Specifically, based on a deep reinforcement learning model (using a multilayer perceptron convolutional neural network), the state space is used as the input to the neural network. It is trained using a large amount of simulated fire scenario data. During training, a reward function is used (e.g., setting reward values ​​based on fire extinguishing efficiency, the degree of protection of tunnel facilities, etc.). Deep reinforcement learning algorithms (such as DQN, A3C, etc.) are used to continuously optimize the parameters of the neural network, enabling it to learn to select the optimal action to control and extinguish the fire under different fire conditions. The parameters of the deep reinforcement learning model are initialized, and the hyperparameters required for training, such as the learning rate, discount factor, and experience replay buffer size, are set. The current environmental state is fed into the neural network for processing, and the neural network outputs the value or probability distribution of each possible action in a given state. Through interaction with the environment, a deep reinforcement learning model capable of making intelligent decisions based on the specific circumstances of a tunnel fire is obtained.

[0045] By combining simulation software with the actual environment of the tunnel, a virtual environment for simulating tunnel fires is constructed. In this virtual environment, the deep reinforcement learning model is optimized through continuous interaction of state-action-reward loops.

[0046] Specifically, a 3D geometric model of the tunnel is created in the selected simulation software. The tunnel's length, width, and height are set according to actual dimensions, and material properties (such as thermal conductivity and fire resistance) are defined. The software's physics engine is used to configure dynamic fire characteristics, simulating the flame spread process by setting the fire source location, heat release rate, and type of combustible material. Temperature field and smoke diffusion modules are added, using computational fluid dynamics (CFD) to simulate hot air flow and smoke concentration distribution within the tunnel. Distributed optical fibers are integrated for real-time acquisition of temperature and vibration data in the virtual environment. The entire virtual environment is verified and debugged to ensure that fire behavior is consistent with physical laws, and virtual environment parameters are adjusted to improve simulation accuracy, thus creating a virtual environment that dynamically reflects the characteristics of tunnel fires. Pre-defined state and action spaces are deployed in the virtual environment, enabling deep reinforcement learning models to interact with them. After the simulation starts, the firefighting robot executes selected actions based on its current state and receives real-time feedback from the fire environment, including new state information and corresponding reward values ​​(calculated based on factors such as fire extinguishing efficiency and energy consumption). Data from the state-action-reward cycle is collected for subsequent analysis. The deep reinforcement learning model is trained using data from the state-action-reward cycle. The parameters of the deep reinforcement learning model are adjusted through batch updates or online learning. This process is repeated continuously, so that the deep reinforcement learning model can learn the optimal behavior pattern after a large number of iterations in the virtual environment.

[0047] The state space is input into the optimized deep reinforcement learning model to generate the movement path of the fire-fighting robot and the start / stop control commands for the water pump.

[0048] Specifically, the current state space data (including the location of the fire source, tunnel environmental parameters, and the status of fire-fighting equipment) is input into an optimized deep reinforcement learning model. Based on the input state information, the deep reinforcement learning model uses its internal neural network to calculate the value or probability distribution of each possible action and selects the optimal action as the output. For a fire-fighting robot, the output action includes specific movement path points or directional commands (such as moving forward or turning left), thereby generating the optimal path from the current location to the target fire source; for a water pump, the output action is a start / stop command or a water pressure adjustment command.

[0049] S4. Control the fire-fighting robot to move according to the fire-fighting robot path and water pump start / stop commands, adjust water pump spray parameters, and collect fire spread data in real time to generate feedback signals.

[0050] The fire-fighting robot moves from its current location to the fire source along a planned path. When the fire-fighting robot approaches the fire source, it starts the water pump and adjusts the spray parameters according to the start and stop control commands of the water pump.

[0051] Specifically, the firefighting robot moves from its current location towards the fire source according to a planned path, continuously monitoring the distance between itself and the fire source. When the robot approaches the fire source and falls within a preset distance threshold (the most effective firefighting distance is between 10 and 20 meters from the fire source, and considering safety and operational flexibility, a distance threshold of 15 meters can be selected as the standard for triggering the water pump to start), it triggers the water pump start command. Furthermore, the robot proceeds along a pre-calculated optimal path, using its navigation system (such as LiDAR, GPS, or a visual recognition system) to correct its direction and position in real time, ensuring precise arrival at the target location. As it approaches the fire source, the deep reinforcement learning model sends a water pump start command based on the latest state space data, simultaneously determining appropriate spray parameters, such as water pressure or spray angle. Upon receiving the water pump start command, the water pump control system immediately responds, adjusting to the specified operating parameters to effectively control the fire.

[0052] Data on fire spread is collected via distributed optical fiber, and the degree of fire reduction and extinguishing efficiency during the firefighting process are recorded and analyzed to evaluate the effectiveness of the firefighting strategy. The evaluation results are then combined with the operating status of the firefighting robot and the working status of the water pump as feedback signals.

[0053] Specifically, distributed optical fibers are used to collect real-time data on the spread of fire within the tunnel, including indicators such as temperature changes and smoke concentration. The reduction in fire intensity and extinguishing efficiency are recorded from the start to the end of firefighting efforts. During the firefighting process, the distributed optical fibers continuously monitor and transmit data to the central processor. For a medium-sized tunnel fire scenario, the initial rapid response phase analyzes the fire intensity every 30 seconds. As the firefighting efforts progress, if a significant reduction in fire intensity is observed, the time interval is adjusted to every 5 minutes. Before confirming that the fire has been completely extinguished, a final check is performed every 1 minute. This involves collecting data every minute on temperature distribution, smoke concentration, the location of the firefighting robot, its battery level, remaining extinguishing agent, and pump status at monitoring points within the tunnel. Based on preset time intervals, changes in fire intensity are analyzed. Standardized evaluation indicators are constructed using parameters such as heat release rate (HRR), temperature gradient, smoke concentration, and flame area. A weighted formula is used to calculate the extinguishing effect value, and time-series trend analysis is combined with judgments (e.g., HRR drops to 10% of the initial value and temperature drops to a safe range) to dynamically assess the extinguishing effect. Key performance indicators (KPIs) such as the rate of temperature decrease in the fire source area, the percentage reduction in smoke concentration, and the required extinguishing time are calculated based on the fire extinguishing effect to quantify the effectiveness of the fire extinguishing strategy. Supplementary information is collected from the operational status of the fire-fighting robot (e.g., movement speed, path accuracy) and the operating status of the water pump (e.g., number of starts, operating duration, water pressure adjustment frequency). The fire extinguishing effect is evaluated in conjunction with this supplementary information to obtain feedback signals.

[0054] It should be noted that the key performance indicators of fire extinguishing effectiveness include the rate of temperature drop in the fire source area, the percentage reduction in smoke concentration, the required extinguishing time, and the amount and efficiency of fire water used. The rate of temperature drop in the fire source area reflects the efficiency of fire extinguishing measures in reducing the temperature at the fire scene. The percentage reduction in smoke concentration assesses the effect of fire extinguishing actions on improving air quality. A shorter required extinguishing time usually indicates a more efficient fire extinguishing strategy.

[0055] S5. Update the reinforcement learning model parameters based on the feedback signals, dynamically optimize the fire robot path and water pump start / stop commands, and re-execute them.

[0056] Feedback signals are input into the deep reinforcement learning model, and the model is trained online using these signals. The parameters of the deep reinforcement learning model are then adjusted using the policy gradient method.

[0057] Specifically, the feedback signals collected during the firefighting process (including the degree of fire reduction, firefighting efficiency, the operating status of the firefighting robot, and the working status of the water pump, etc.) are input into the deep reinforcement learning model as new training data. The reward value of the current strategy is calculated using the feedback signals, and the quality of the action taken by the deep reinforcement learning model in the current state is evaluated based on the reward value. The Policy Gradient Approach (PPO algorithm) is used to calculate the gradient of the loss function based on the relationship between reward values ​​and action probability distributions. Gradient ascent is then used to update the parameters of the deep reinforcement learning model to optimize the policy. A batch of data containing states, actions, rewards, and the next state is sampled from the experience replay buffer. The advantage function value for each action is calculated, and the weight parameters of the neural network are adjusted based on the advantage function value. This process is repeated, allowing the deep reinforcement learning model to learn online and gradually improve its decision-making ability.

[0058] Based on the updated deep reinforcement learning model, the optimized movement path of the fire-fighting robot and the start / stop control commands of the water pump are recalculated. The fire-fighting robot readjusts its path according to the optimized path and readjusts its spray parameters according to the optimized start / stop control commands of the water pump after approaching the fire source.

[0059] Specifically, the current environmental conditions (including the location of the fire source, tunnel conditions, and the status of fire-fighting equipment) are input into the deep reinforcement learning model to recalculate and optimize the movement path of the fire-fighting robot and the start / stop control commands for the water pumps. Based on the new parameters and strategies, the deep reinforcement learning model generates a more efficient path to the fire source and precise water pump operation instructions. The fire-fighting robot receives the optimized path and adjusts its route in real time through navigation, accurately moving towards the fire source. When the robot approaches the fire source and reaches a preset distance, it starts or adjusts the water pumps according to the optimized water pump control commands, such as adjusting parameters like spray pressure or angle, to perform fire-fighting operations. Throughout the process, the robot continuously monitors changes in the surrounding environment and feeds new data back to the deep reinforcement learning model. When necessary, it can further optimize the path and water pump control strategies, thereby achieving dynamic response and efficient fire-fighting.

[0060] S6. Obtain the fire safety threshold based on the tunnel cross-sectional area, assess the fire status using the fire safety threshold, and initiate multispectral scanning to detect the fire extinguishing situation.

[0061] Based on the tunnel's geometry, the airflow capacity and heat diffusion capacity are calculated. Then, based on the tunnel parameters, fire dynamics theory, and tunnel cross-sectional area, the fire safety threshold is obtained by combining the airflow capacity and heat diffusion capacity.

[0062] Specifically, based on the tunnel's geometry, including parameters such as tunnel length, cross-sectional area, longitudinal slope, and ventilation shaft spacing, and combined with tunnel ventilation design data, the airflow rate through the tunnel per unit time is calculated by measuring the wind speed at the tunnel entrance, the pressure difference at the exit, and the power of the ventilation equipment, thus obtaining the air circulation capacity. Simultaneously, based on fire dynamics theory, and combined with the tunnel's cross-sectional area and data from past fire cases, the heat diffusion capacity during a fire is calculated, with a focus on analyzing the impact of fire source power, type of combustible material, and ventilation conditions on heat propagation. Based on the ratio of tunnel cross-sectional area to air circulation capacity, the safety factor is dynamically adjusted. When the tunnel cross-sectional area is greater than 150m² or the ventilation capacity is less than 120,000m³ / h, the safety factor is increased to 1.5. When the tunnel cross-sectional area is less than or equal to 100m² and the ventilation capacity is greater than or equal to 180,000m³ / h, the safety factor is decreased to 0.8. Taking into account air circulation capacity, heat diffusion capacity, and safety factor, and combined with real-time monitoring data from temperature sensors inside the tunnel, the fire safety threshold is set at 800℃.

[0063] Distributed optical fiber is used to monitor temperature distribution, smoke concentration, and flame coverage within the tunnel. The collected fire data is compared with fire safety thresholds to assess the fire status. Specifically, a distributed fiber optic network is used to monitor the temperature distribution, smoke concentration, and flame coverage within the tunnel in real time. The distributed fiber optic cables, laid along the tunnel, provide a continuous data stream, transmitting the collected real-time data (such as temperature rise in specific areas, smoke concentration levels, and the length or area affected by the flames) to a data analysis system. The real-time monitoring data is compared with pre-calculated fire safety thresholds, which include indicators such as maximum allowable temperature rise and smoke concentration limits. Based on the comparison results, the current fire status is assessed: if any one or more parameters exceed the safety threshold, the fire has reached a dangerous level and immediate action is required; if all parameters are within safe limits, the fire is considered controllable.

[0064] Based on the fire situation, a multispectral scan is initiated to comprehensively scan the fire area and detect the fire suppression progress.

[0065] Specifically, based on the fire status (such as temperature distribution, smoke concentration, and flame coverage) monitored by distributed optical fibers, it is determined whether to activate the multispectral scanning equipment; if the fire is close to or exceeds the safety threshold, the multispectral scanning equipment is activated immediately. A comprehensive scan of the fire area is performed, using multi-band spectral signals (such as visible light, infrared, and near-infrared) to obtain information such as the thermal radiation distribution, location of residual hotspots, and smoke dissipation in the fire area. Image processing and spectral feature extraction techniques are used to detect the presence of open flames or high-temperature areas and to assess the effectiveness of fire suppression: if residual flames or abnormally high-temperature points are found, their specific locations are marked and it is determined that the fire has not been completely extinguished; if all fire area indicators meet safety standards, it is confirmed that the fire has been effectively controlled.

[0066] This embodiment also provides an intelligent monitoring and operation system for fire fighting in tunnel highways, including: a data acquisition module, a positioning module, an instruction generation module, an execution module, an optimization module, and a detection module; The acquisition module is used to acquire vibration signals through distributed optical fibers on the tunnel sidewalls, and perform wavelet noise reduction processing to extract vibration features and obtain vibration feature vectors. The localization module is used to input vibration feature vectors into a pre-trained fire classification model to obtain fire events, and to calculate the location of the fire source through time-domain reflectometry. The instruction generation module is used to build a deep reinforcement learning model based on the fire source location and tunnel state parameters, and generate fire robot paths and water pump start / stop instructions. The execution module is used to control the fire-fighting robot to move according to the fire-fighting robot path and water pump start / stop commands, adjust water pump spray parameters, and collect fire spread data in real time to generate feedback signals. The optimization module is used to update the reinforcement learning model parameters based on feedback signals, dynamically optimize the fire robot path and water pump start / stop commands, and re-execute them. The detection module is used to set a fire safety threshold based on the tunnel cross-sectional area, assess the fire status through the fire safety threshold, and detect the fire extinguishing status.

[0067] This embodiment also provides a computer device for an intelligent monitoring and operation method for fire prevention in tunnels and highways, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the intelligent monitoring and operation method for fire prevention in tunnels and highways as proposed in the above embodiment.

[0068] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0069] This embodiment also provides a storage medium storing a computer program. When executed by a processor, the program implements the intelligent monitoring and operation method for tunnel and highway fire fighting as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0070] In summary, this invention achieves highly sensitive monitoring and early warning of fire hazards by using distributed optical fibers to collect vibration signals on the tunnel sidewalls and extracting precise vibration feature vectors through wavelet denoising processing. This effectively reduces the false alarm rate and improves the accuracy of fire source location. Furthermore, by utilizing time-domain reflectometry to accurately locate the fire source, it provides crucial information and significantly enhances the intelligence level and emergency response capability of tunnel highway fire prevention and control.

[0071] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An intelligent monitoring and operation method for fire fighting in tunnel highways, characterized in that: include, Vibration signals are collected by distributed optical fibers on the tunnel sidewalls, and wavelet noise reduction is performed to extract vibration features, resulting in a vibration feature vector. The vibration feature vector is input into a pre-trained fire classification model to obtain fire events, and the location of the fire source is calculated by temporal reflectance localization. A deep reinforcement learning model is constructed based on the fire source location and tunnel state parameters to generate fire robot paths and water pump start / stop commands. The system controls the fire-fighting robot to move according to the fire-fighting robot path and water pump start / stop commands, and adjusts the water pump spray parameters, while collecting real-time fire spread data and generating feedback signals. The parameters of the reinforcement learning model are updated based on feedback signals, and the fire robot path and water pump start / stop commands are dynamically optimized and re-executed. Fire safety thresholds are set based on the tunnel cross-sectional area, and the fire status is assessed and the fire extinguishing status is monitored using these fire safety thresholds. The vibration signal is collected through distributed optical fibers on the tunnel sidewall, and wavelet noise reduction processing is performed to extract vibration features, resulting in a vibration feature vector. The specific steps are as follows: Vibration signals are generated by collecting data on the tunnel's structural health, external influences, and traffic load through distributed optical fibers along the tunnel sidewalls. Noise and interference signals in the vibration signal are removed by filtering, and the vibration signal is analyzed by wavelet transform to decompose the vibration signal into different frequency components. The vibration signal is then reconstructed by inverse wavelet transform to obtain the noise-reduced vibration signal. The vibration signal after noise reduction is analyzed by wavelet transform, and the time domain, frequency domain and wavelet energy distribution features are extracted. The extracted features are combined into a feature vector to obtain the vibration feature vector. The process of inputting vibration feature vectors into a pre-trained fire classification model to obtain fire events, and calculating the location of the fire source through time-domain reflectometry, involves the following specific steps. The vibration feature vector is input into a pre-trained fire classification model for analysis to obtain the fire distribution probability and output the category label. Based on the category label, the time-domain reflection positioning function of the distributed optical fiber is triggered. By sending pulse signals to the distributed optical fiber and receiving reflected signals, the specific location of the fire source on the optical fiber is calculated based on the signal propagation time and reflection characteristics. Combined with the deployment path of the distributed optical fiber on the tunnel sidewall, the location of the fire source is mapped to the actual tunnel spatial coordinates.

2. The intelligent monitoring and operation method for fire fighting in tunnel highways as described in claim 1, characterized in that: The process involves constructing a deep reinforcement learning model based on the fire source location and tunnel state parameters to generate the fire robot's path and water pump start / stop commands. The specific steps are as follows: The state space is composed of the fire source location, tunnel status parameters and fire equipment status, and the action space is defined according to the activity range of the fire robot and the switching action of the water pump. By combining deep reinforcement learning algorithms with neural networks, a deep reinforcement learning model is constructed based on the state space and action space. By combining simulation software with the actual environment of the tunnel, a virtual environment for simulating tunnel fires is constructed. In the virtual environment, the deep reinforcement learning model is optimized through continuous interaction of state-action-reward loop. The state space is input into the optimized deep reinforcement learning model to generate the movement path of the fire-fighting robot and the start / stop control commands for the water pump.

3. The intelligent monitoring and operation method for fire fighting in tunnel highways as described in claim 2, characterized in that: The fire-fighting robot is controlled to move according to its path and water pump start / stop commands, and adjusts the water pump spray parameters. It also collects real-time fire spread data to generate feedback signals. The specific steps are as follows: The fire-fighting robot moves from its current location to the fire source according to the planned path. When the fire-fighting robot approaches the fire source, it starts the water pump and adjusts the water pump spray parameters according to the start and stop control commands of the water pump. Data on fire spread is collected via distributed optical fiber, and the degree of fire reduction and extinguishing efficiency during the firefighting process are recorded and analyzed to evaluate the effectiveness of the firefighting strategy. The evaluation results are then combined with the operating status of the firefighting robot and the working status of the water pump as feedback signals.

4. The intelligent monitoring and operation method for fire fighting in tunnel highways as described in claim 3, characterized in that: The steps for updating the reinforcement learning model parameters based on feedback signals, dynamically optimizing the fire robot path and water pump start / stop commands, and re-executing them are as follows: The feedback signal is input into the deep reinforcement learning model, and the deep reinforcement learning model is learned online using the feedback signal. The parameters of the deep reinforcement learning model are adjusted by the policy gradient method. Based on the updated deep reinforcement learning model, the optimized movement path of the fire-fighting robot and the start / stop command of the water pump are recalculated. The fire-fighting robot readjusts its movement path according to the optimized path and readjusts the spray parameters according to the optimized start / stop control command of the water pump after approaching the fire source.

5. The intelligent monitoring and operation method for fire fighting in tunnel highways as described in claim 4, characterized in that: The fire safety threshold is obtained based on the tunnel cross-sectional area, and the fire status is assessed using this threshold. Multispectral scanning is then initiated to detect the fire suppression progress. The specific steps are as follows: Based on the tunnel's geometry, the air circulation capacity and heat diffusion capacity are calculated. The fire safety threshold is obtained by combining the tunnel parameters, fire dynamics theory, and tunnel cross-sectional area with the air circulation capacity and heat diffusion capacity. Distributed optical fiber is used to monitor temperature distribution, smoke concentration, and flame coverage within the tunnel. The collected data is compared with fire safety thresholds to assess the fire status. Based on the fire situation, a multispectral scan is initiated to comprehensively scan the fire area and detect the fire suppression progress.

6. An intelligent monitoring and operation system for fire fighting in tunnel highways, based on the intelligent monitoring and operation method for fire fighting in tunnel highways as described in any one of claims 1 to 5, characterized in that: It includes a data acquisition module, a positioning module, an instruction generation module, an execution module, an optimization module, and a detection module; The acquisition module is used to acquire vibration signals through distributed optical fibers on the tunnel sidewall, and perform wavelet noise reduction processing to extract vibration features and obtain vibration feature vectors. The positioning module is used to input the vibration feature vector into a pre-trained fire classification model to obtain fire events, and to calculate the location of the fire source through time-domain reflection positioning. The instruction generation module is used to construct a deep reinforcement learning model based on the fire source location and tunnel state parameters, and generate fire robot paths and water pump start / stop instructions. The execution module is used to control the fire-fighting robot to move according to the fire-fighting robot path and water pump start / stop commands, and to adjust the water pump spray parameters, and to collect fire spread data in real time to generate feedback signals. The optimization module is used to update the reinforcement learning model parameters according to the feedback signal, dynamically optimize the fire robot path and water pump start / stop commands, and re-execute them; The detection module is used to set a fire safety threshold based on the tunnel cross-sectional area, assess the fire status through the fire safety threshold, and detect the fire extinguishing status.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the intelligent monitoring and operation method for fire fighting in tunnels and highways as described in any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the intelligent monitoring and operation method for fire fighting in tunnels and highways as described in any one of claims 1 to 5.

Images