Fire-fighting linkage control device for petroleum and petrochemical engineering

Abstract

The invention relates to the technical field of fire-fighting control, in particular to a fire-fighting linkage control device for petroleum and petrochemical engineering, which comprises a multi-modal data acquisition module, a fire modeling and prediction module, a self-adaptive fire extinguishing control module and a virtual reproduction and optimization module. Dynamically modeling a flame form through multi-modal data fusion and a generative adversarial network; a fire path is accurately predicted in combination with a differential equation network; a path planning algorithm is adopted to optimize equipment scheduling, and modeling parameters and equipment configuration are analyzed and optimized through virtual scene generation and feedback data. According to the method, the accuracy of fire scene modeling and the efficiency of fire extinguishing equipment dispatching are remarkably improved, the problems that a traditional fire extinguishing system is slow in response and low in control accuracy are solved, and a powerful technical guarantee is provided for safe operation of petroleum and petrochemical storage tanks.

Description

Technical Field

[0001] The invention relates to the technical field of fire control, and in particular to a fire linkage control device for petrochemical industry. Background Art

[0002] In the petroleum and petrochemical industry, fire safety in the tank area is extremely important. Once a fire occurs in a tank, it is very easy to cause a large-scale explosion and environmental pollution due to the complex types of combustible materials, rapid fire spread, and high thermal radiation intensity, resulting in huge economic losses and safety risks. At present, traditional fire control devices mostly rely on manual operation, mainly through on-site manual control of fire pumps, valves and alarms to achieve fire emergency.

[0003] However, the existing technology has the following shortcomings: it relies on manual judgment of the fire situation and cannot respond quickly to changes in the fire; the existing systems mostly operate equipment through fixed procedures, lack dynamic adjustment capabilities, and find it difficult to accurately respond to complex fire scenarios; there is a lack of real-time modeling and prediction of flame morphology, diffusion paths, and environmental variables, resulting in blind fire-fighting strategies and inefficient use of fire-fighting resources. Summary of the invention

[0004] In view of the many problems existing in the above-mentioned prior art, the present invention provides a fire linkage control device for petrochemical industry. The present invention is based on multimodal data fusion technology, realizes dynamic modeling of flame morphology through generative adversarial network, and accurately predicts the fire spread path in combination with differential equation network. The adaptive fire extinguishing control module optimizes equipment scheduling by using real-time path planning algorithm, and dynamically adjusts modeling parameters and equipment configuration in combination with virtual reproduction and optimization module, thereby significantly improving fire control efficiency and response accuracy.

[0005] A fire linkage control device for petrochemical industry, comprising: The multimodal data acquisition module includes a variety of sensors for collecting thermal data, smoke particle data, and vibration data; it is used to fuse the collected data to generate initial fire scene data; The fire modeling and prediction module is used to receive the initial fire scene data, perform flame morphology modeling through the generative adversarial network, and perform path prediction through the differential equation network to generate fire dynamic prediction data; Adaptive fire extinguishing control module, which is used to allocate fire extinguishing tasks and optimize fire extinguishing paths according to fire dynamic prediction data. The optimized fire extinguishing path uses a path planning algorithm combined with a real-time adjustment mechanism to adjust the operating parameters of the fire extinguishing equipment; The virtual reproduction and optimization module is used to receive feedback data of the fire extinguishing process and generate a virtual scene for virtual reproduction of the fire scene. The virtual scene is used to train operators and optimize equipment configuration, and optimize the modeling parameters of the adversarial network and the path prediction parameters of the differential equation network based on the feedback data.

[0006] Preferably, the multimodal data acquisition module further includes an environmental humidity sensor and a wind speed sensor, and generates initial fire scene data by fusing the environmental humidity data and wind speed data with thermal temperature data, smoke particle data and vibration data.

[0007] Preferably, the mimetic neural network of the multimodal data acquisition module converts thermal data, smoke particle data, vibration data, ambient humidity data and wind speed data into a unified comprehensive feature vector by establishing a correlation model between time series features and spatial distribution features. The comprehensive feature vector is optimized by a graph Laplace dimensionality reduction algorithm to generate initial fire scene data.

[0008] Preferably, the generator of the generative adversarial network of the fire modeling and prediction module generates flame morphology data based on the input initial fire scene data, and the discriminator calculates the error between the generated flame morphology data and the real flame morphology data, and feeds the error back to the generator to adjust its network parameters until the generated flame morphology data matches the real flame morphology data.

[0009] Preferably, the differential equation network of the fire modeling and prediction module calculates the fire spread path by analyzing the physical field parameters of the heat conduction characteristics, the combustion reaction rate and the fire spread direction, and dynamically adjusts the path prediction results to cope with the real-time changes of environmental variables.

[0010] Preferably, the task allocation unit of the adaptive fire extinguishing control module allocates tasks to drones and fire extinguishing robots based on fire dynamic prediction data, and sets the execution order according to the importance level of the task area to ensure maximum resource utilization and improved fire control efficiency.

[0011] Preferably, the path optimization unit of the adaptive fire extinguishing control module generates an optimal moving path by using a path planning algorithm based on a cost function by updating the obstacle distribution in the fire area, the current position of the equipment and the shortest path cost in real time, thereby ensuring that the fire extinguishing equipment reaches the task point efficiently.

[0012] Preferably, the holographic scene reconstruction unit of the virtual reproduction and optimization module generates a dynamic scene including flame height, spreading speed and equipment position by performing three-dimensional reconstruction on multimodal data of the fire scene.

[0013] Preferably, the virtual reproduction and optimization module superimposes the holographic scene into the operator's field of view through an augmented reality auxiliary system to display the real-time spread path of the fire, the operating status of the fire extinguishing equipment and changes in the surrounding environment.

[0014] Preferably, the virtual reproduction and optimization module adjusts the generator learning rate and discriminator error function parameters of the generative adversarial network in real time by analyzing the feedback data of the fire extinguishing process, and optimizes the physical field parameter calculation formula in the differential equation network.

[0015] Compared with the prior art, the advantages and beneficial effects of the present invention are: The present invention uses multimodal data fusion technology to achieve high-precision dynamic modeling of fire scenes, significantly improving the accuracy of flame spread path prediction; The present invention realizes the real-time optimization of fire morphology modeling and path prediction by combining generative adversarial networks with differential equation networks, effectively making up for the defect that traditional systems are insufficiently responsive to dynamic changes in fires. The present invention provides operator training and fire extinguishing equipment configuration optimization functions through virtual reproduction and optimization modules, which significantly improves the emergency response capability and fire extinguishing efficiency of the fire protection system; The present invention optimizes equipment scheduling and path planning through an adaptive fire extinguishing control module, maximizes the utilization efficiency of fire extinguishing resources, and effectively solves the problem of rigidity of traditional equipment scheduling methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a structural block diagram of the device of the present invention; Figure 2 It is a schematic diagram of the multimodal data fusion and fire modeling process in the present invention; Figure 3 It is a schematic diagram of the interaction and dynamic adjustment process of the virtual reproduction and optimization modules in the present invention; Figure 4 It is a schematic diagram of the adaptive fire extinguishing control module and path optimization in the present invention. DETAILED DESCRIPTION

[0017] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present disclosure. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is obvious that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0018] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise", "include", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0019] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0020] like Figure 1-2 As shown, a fire linkage control device for petrochemical industry comprises: The multimodal data acquisition module includes a variety of sensors for collecting thermal data, smoke particle data, and vibration data; it is used to fuse the collected data to generate initial fire scene data; Preferably, the multimodal data acquisition module further includes an environmental humidity sensor and a wind speed sensor, and generates initial fire scene data by fusing the environmental humidity data and wind speed data with thermal temperature data, smoke particle data and vibration data.

[0021] Thermal data is obtained through a high dynamic range thermal imaging sensor, which can capture the temperature distribution in the tank area and the thermal radiation characteristics of the flame, generating two-dimensional temperature field data. Thermal data is an important input for fire modeling, used to identify high temperature areas, infer the location of the fire source and its spread trend.

[0022] Smoke particle data is obtained through a laser scattering analysis system. The scattering intensity of the laser beam in the smoke is proportional to the particle concentration. By measuring the change in the scattering signal, the particle concentration distribution and diffusion direction are calculated. Smoke particle data is used to predict the spread path of the fire and assist in judging the interaction between the flame and obstacles.

[0023] Vibration data is collected by an array of piezoelectric microsensors mounted on the surface of the tank, which can detect heat conduction and combustion vibrations caused by the fire. Through fast Fourier transform (FFT), the vibration signal is converted into frequency domain features for analyzing the combustion intensity and heat transfer path of the fire.

[0024] Ambient humidity data is measured by humidity sensors, and wind speed and direction data are obtained by wind speed sensors. These two types of data are crucial in fire modeling and can be used to correct the calculation results of the flame spread path. A high humidity environment may slow down the spread of flames, while wind speed and direction directly affect the speed and direction of fire spread.

[0025] All collected data is fused and processed through a mimetic neural network. Mimetic neural networks imitate the working mechanism of biological neural networks and can dynamically learn and associate the features of multimodal data to generate unified initial fire scene data. Specifically, each layer in the network extracts features from a single sensor data, and then aligns the features of multiple sensor data through a cross-modal connection layer, and finally generates a comprehensive feature vector as the basic data for fire modeling.

[0026] Through the collection and fusion processing of the above multimodal data, the actual scene characteristics of the fire can be accurately reflected, providing high-quality input data for fire dynamic modeling and prediction. Combined with environmental humidity and wind speed data, the accuracy of fire spread path calculation can be further improved, effectively supporting the dispatch and strategy optimization of fire extinguishing equipment.

[0027] Preferably, the mimetic neural network of the multimodal data acquisition module converts thermal data, smoke particle data, vibration data, ambient humidity data and wind speed data into a unified comprehensive feature vector by establishing a correlation model between time series features and spatial distribution features. The comprehensive feature vector is optimized by a graph Laplace dimensionality reduction algorithm to generate initial fire scene data.

[0028] The simulated neural network simulates the excitation and transmission mechanism of biological neurons and has the ability to dynamically learn and associate data features of different modalities. In the present invention, the processing flow of the simulated neural network is as follows: Through the first layer of the network, preliminary feature extraction is performed on five types of data, such as thermal data and smoke particle data, to generate a single-mode time series feature vector and a spatial distribution feature vector; the middle layer establishes an association model between time series features and spatial distribution features, integrates the correlation between different modal data, and generates a multimodal feature map; the output layer of the network converts the multimodal feature map into a unified comprehensive feature vector, which contains the global dynamic information of the fire scene, including key parameters such as temperature gradient, smoke diffusion direction, and wind speed influence.

[0029] Since the dimension of the comprehensive feature vector is high, in order to improve the computational efficiency, the scheme introduces the graph Laplace dimensionality reduction algorithm to reduce the dimension of the data. The algorithm builds a graph structure between data points, retains the core geometric features of high-dimensional data, reduces redundant information, and generates an optimized low-dimensional feature vector.

[0030] The core optimization formula is as follows: , in, Represents the low-dimensional feature vector after dimensionality reduction; , Represent data points in a high-dimensional feature space; represents the weight coefficient, represents the data point With data points The similarities between.

[0031] By minimizing the above formula, the most important part of the feature vector for fire modeling and prediction is retained.

[0032] Through the optimization of multimodal data fusion of the pseudo-neural network and the graph Laplace dimension reduction algorithm, the generated initial fire scene data is highly accurate and complete. This data fully reflects the dynamic characteristics of the fire scene, can significantly improve the accuracy of subsequent fire modeling and path prediction, and greatly reduce data processing time, thereby improving the response speed and execution efficiency of fire fighting decisions.

[0033] Example: In the emergency response to a fire in a petrochemical tank area, the on-site environment is complex, the flames spread rapidly with smoke, and the wind speed is high and the direction is changeable. In order to ensure the scientific nature of the fire extinguishing strategy, it is necessary to quickly integrate the multimodal sensor data at the scene and generate high-quality initial fire scene data.

[0034] The thermal imaging sensor collects the temperature distribution of the flame area and generates a thermal temperature data feature vector. The laser scattering analysis system records the concentration and diffusion rate of smoke particles and generates a smoke particle data feature vector. The piezoelectric microsensor detects the combustion vibration signal and generates a vibration data feature vector through fast Fourier transform (FFT). The humidity sensor and wind speed sensor record the ambient humidity and wind speed direction respectively, generating humidity feature data and wind speed feature data.

[0035] The five types of data are input into the feature extraction layer of the simulated neural network to generate time series features and spatial distribution feature vectors. The middle layer of the network establishes an association model between time series features and spatial distribution features to generate a multimodal feature map. The output layer converts the multimodal feature map into a comprehensive feature vector, which contains the global dynamic features of the fire scene. The comprehensive feature vector is optimized by the graph Laplace dimensionality reduction algorithm to generate a low-dimensional feature vector for constructing the initial fire scene data.

[0036] The initial fire scene data generated through the above process accurately reflects the core information such as the temperature distribution of the flame, the smoke diffusion path, the direction of wind speed influence, etc. Combined with these data, the system successfully predicted the spread trend of the fire, and optimized the dispatching strategy and path planning of the fire extinguishing equipment accordingly, significantly reducing the control time of the fire.

[0037] The fire modeling and prediction module is used to receive the initial fire scene data, perform flame morphology modeling through the generative adversarial network, and perform path prediction through the differential equation network to generate fire dynamic prediction data; The fire modeling and prediction module is the core component of the petroleum and petrochemical fire linkage control device. It receives the initial fire scene data generated by the multimodal data acquisition module, models the flame morphology through the generative adversarial network (GAN), and uses the differential equation network (ODE-Net) to dynamically predict the fire spread path. It finally generates fire dynamic prediction data to provide accurate support for the formulation of fire extinguishing strategies and equipment scheduling.

[0038] Generative Adversarial Networks (GANs) consist of a generator and a discriminator.

[0039] Based on the input initial fire scene data, the generator simulates the dynamic morphology of flames under different environmental conditions and generates a high-resolution flame image sequence that reflects the flame diffusion trend and combustion intensity in the fire.

[0040] The discriminator calculates the output error of the generator by comparing the generated flame image with the actual fire data, and feeds the error back to the generator to adjust the network parameters.

[0041] Through continuous iterative training, the generator can generate flame morphology data that is highly matched with real fire scenes.

[0042] The differential equation network (ODE-Net) is used to simulate the dynamic propagation path of fire. Based on the heat conduction, air flow and combustion reaction rate of the fire, a dynamic model of flame diffusion is constructed. ODE-Net solves the above differential equations through numerical methods and calculates the propagation path of the flame in different time and space.

[0043] By generating adversarial networks to model flame morphology, we can accurately reproduce the flame morphology at the fire scene, capture high-temperature areas and flame boundaries, and provide accurate input data for subsequent fire path prediction. The application of differential equation networks can dynamically simulate the flame diffusion path and the direction and speed of fire spread, providing a reliable basis for the scheduling and deployment of fire extinguishing equipment. The combination of the two significantly improves the accuracy and real-time performance of fire dynamic prediction, thereby optimizing fire extinguishing strategies and improving fire control efficiency.

[0044] Preferably, the generator of the generative adversarial network of the fire modeling and prediction module generates flame morphology data based on the input initial fire scene data, and the discriminator calculates the error between the generated flame morphology data and the real flame morphology data, and feeds the error back to the generator to adjust its network parameters until the generated flame morphology data matches the real flame morphology data.

[0045] In the petroleum and petrochemical fire linkage control device, the fire modeling and prediction module uses the generative adversarial network (GAN) to dynamically model the flame morphology. GAN consists of a generator and a discriminator, which work together in an adversarial manner to ultimately generate flame morphology data that is close to the real fire scene, providing key input for subsequent fire path prediction and fire extinguishing strategy optimization.

[0046] The generator receives the initial fire scene data (including thermal data, smoke particle data, etc.) and generates flame morphology data through a deep neural network. These data are presented in the form of flame images or heat distribution maps, describing the height, shape, and burning intensity of the flame. The goal of the generator is to gradually approximate the real flame morphology so that the flame images it generates are difficult to distinguish from the flame images collected in the real fire scene.

[0047] The task of the discriminator is to distinguish the flame morphology data output by the generator from the real flame data. By calculating the error between the two, the discriminator outputs a authenticity score. During the GAN training process, the discriminator's goal is to improve the ability to distinguish between real data and generated data, thereby forcing the generator to continuously improve the generated results.

[0048] The generator and the discriminator optimize their performance through alternating training. The loss function is used to evaluate the performance of the generator and the discriminator. Its core formula is: , in, represents the loss function; It represents the discriminant result of the real flame data; Represents the flame shape data generated by the generator; Represents the distribution of real flame data; represents the input noise distribution of the generator.

[0049] During the training process, the discriminator feeds back the error to the generator, and the generator adjusts the network parameters to make the generated results approach the real flame data. The training is continued until the generated flame shape data is highly matched with the real data.

[0050] Through the modeling of generative adversarial networks, the system can dynamically generate flame morphology data that conforms to the characteristics of real fires. These data fully reflect the height, diffusion range and burning intensity of the flames, providing high-quality input for subsequent fire path prediction. At the same time, the model has adaptive learning capabilities and can dynamically adjust network parameters according to different fire scenarios to ensure the accuracy and real-time nature of the generated data, thereby improving the decision-making support capabilities of the entire fire linkage device.

[0051] Embodiment, in the fire handling of a petrochemical tank area, the flame height and diffusion speed change rapidly, and it is necessary to generate flame morphology data in real time to accurately assess the scale of the fire and provide support for the deployment of fire extinguishing equipment.

[0052] The initial fire scene data includes thermal data, smoke particle data, vibration data, etc., which are used as input feature vectors of the generator. The generator generates a sequence of flame images based on the input features. These images reflect the dynamic changes of the flame, including flame height, boundary shape, and heat intensity distribution.

[0053] The discriminator compares the flame image generated by the generator with the real flame image collected on site and outputs a authenticity score. The discriminator calculates the error between the generated image and the real image and feeds it back to the generator to adjust its network parameters.

[0054] Through multiple rounds of iterative training, the generator gradually improves the output results, and the difference between the generated flame morphology data and the real flame data gradually decreases. When the loss function converges, the flame image output by the generator meets the requirements of a real fire scene.

[0055] After the above process, the system successfully generated flame morphology data that was highly consistent with the actual situation of the fire on site. These data accurately described the height, diffusion direction and burning intensity of the flames, providing efficient support for fire path prediction and fire extinguishing equipment scheduling. In actual operation, the system optimized the deployment path of the fire extinguishing equipment based on these data, quickly controlled the fire and prevented the fire from spreading further.

[0056] Preferably, the differential equation network of the fire modeling and prediction module calculates the fire spread path by analyzing the physical field parameters of the heat conduction characteristics, the combustion reaction rate and the fire spread direction, and dynamically adjusts the path prediction results to cope with the real-time changes of environmental variables.

[0057] The fire modeling and prediction module in the present invention uses a differential equation network (ODE-Net) to dynamically calculate and predict the fire spread path. By analyzing the heat conduction characteristics, combustion reaction rate and the influence of environmental variables on flame spread in the fire physical field, the differential equation network can simulate the spread behavior of the fire and dynamically adjust the prediction results to adapt to the real-time changing environmental parameters.

[0058] Heat conduction is an important physical mechanism for fire spread. ODE-Net models the temperature change of the tank surface and the surrounding air based on Fourier's law of heat conduction. The heat conduction equation is: , in, represents the temperature field; Indicates time; represents the thermal diffusion coefficient; represents the spatial second-order derivative of temperature, indicating the intensity of heat diffusion. By numerically solving this equation, ODE-Net is able to predict the heat propagation path of the fire on the tank surface and surrounding areas.

[0059] The combustion reaction rate of a fire directly affects the diffusion speed and intensity of the flame. ODE-Net models the combustion reaction rate through nonlinear reaction terms. The combustion equation is: , in, represents the combustion reaction rate; Represents the combustion reaction rate coefficient. This model can describe the dynamic process of flame rapidly spreading from the initial point to the surrounding area.

[0060] Environmental parameters such as wind speed and humidity have a significant impact on the fire spread path. ODE-Net simulates these factors by introducing external disturbance terms. The expanded fire spread model is: , in, represents the wind speed vector field; Represents the external disturbance coefficient. By updating environmental parameters in real time, the network can dynamically adjust the prediction results to ensure the accuracy of fire path prediction.

[0061] Through the differential equation network, the fire modeling and prediction module can calculate the fire spread path in real time and accurately predict the flame spread direction, speed and range. This modeling method based on heat conduction and combustion reaction ensures the physical accuracy of the prediction results. At the same time, the introduction of the dynamic adjustment mechanism of environmental variables makes the model more adaptable and robust. Combined with the data input of the actual fire scene, the system can generate high-precision fire dynamic prediction data in a short time, providing core support for the scheduling and path optimization of fire extinguishing equipment.

[0062] Example: A fire broke out in a petrochemical tank area, and the flames spread rapidly. The surrounding wind speed was high and the humidity was low. In order to effectively control the fire, it is necessary to predict the fire spread path in real time to provide accurate guidance for the deployment of fire-fighting equipment and the evacuation of personnel. Implementation process: The initial fire scenario data includes thermal data of the tank surface, smoke density, vibration signal, ambient humidity and wind speed. These data are input into the differential equation network as the initial conditions of the model.

[0063] The Fourier heat conduction equation is used to simulate the heat diffusion path of the flame and predict the high temperature area on the tank surface. During the numerical solution process, the thermal diffusion coefficient of the tank material is introduced to ensure that the prediction results are consistent with the actual physical properties.

[0064] The combustion reaction rate equation is used to simulate the flame diffusion rate at different combustion stages. Combined with the on-site smoke concentration data, the combustion reaction rate coefficient is corrected to optimize the prediction results of the fire spread rate.

[0065] According to the real-time updated wind speed and humidity data, the external disturbance terms are adjusted to dynamically correct the prediction of the fire spread path. The prediction results show that the flame spreads faster in the direction of higher wind speed, and the flame spread is more significant in the low humidity area.

[0066] Output fire dynamic prediction data, including the location of high-risk areas, the direction and speed of flame spread. Based on the prediction data, the system optimizes the path planning of drones and fire-fighting robots, and concentrates fire-fighting resources on high-risk areas.

[0067] The system successfully predicted the direction of flame spread and high-risk areas, and optimized the dispatch and path planning of fire-fighting equipment. The fire-fighting operation was accurate and rapid, and the fire was effectively controlled in a short period of time, preventing the fire from spreading further to the adjacent tank area, minimizing economic losses and personnel risks.

[0068] Adaptive fire extinguishing control module, which is used to allocate fire extinguishing tasks and optimize fire extinguishing paths according to fire dynamic prediction data. The optimized fire extinguishing path uses a path planning algorithm combined with a real-time adjustment mechanism to adjust the operating parameters of the fire extinguishing equipment; The adaptive fire extinguishing control module is one of the core functional modules of the petroleum and petrochemical fire linkage control device. It is responsible for intelligently allocating fire extinguishing tasks based on fire dynamic prediction data, and optimizing the operation path and parameters of fire extinguishing equipment through path planning algorithms and real-time adjustment mechanisms. This module is designed to ensure that fire extinguishing equipment can complete fire extinguishing tasks efficiently and safely in complex fire environments.

[0069] This module receives fire dynamic prediction data from the fire modeling and prediction module, including flame height, spread path, location of high-risk areas, and environmental parameters (such as wind speed and humidity). These data are analyzed and used to calculate the optimal task allocation plan and path planning strategy for the equipment.

[0070] Task allocation is based on the importance level of the fire area. The system divides the task area into high, medium and low priorities by calculating the risk value of the fire area (such as temperature gradient, flame spread speed, threat to surrounding tanks), and assigns tasks based on the performance of the equipment (such as the mobility of drones and the load capacity of fire-fighting robots). Task allocation follows the following optimization goals: maximize the flame suppression efficiency in high-risk areas; balance the equipment load to avoid overuse of a single device; minimize the time it takes for the equipment to reach the task point.

[0071] Path optimization uses the A* algorithm to find the optimal path from the current location of the equipment to the task point through heuristic search. The real-time adjustment mechanism dynamically optimizes the path planning and equipment operating parameters (such as spray intensity, angle, and moving speed) based on the latest fire dynamics data to ensure that the equipment always moves along the optimal path and achieves the highest fire extinguishing efficiency.

[0072] The adaptive fire extinguishing control module effectively improves the response speed and resource utilization of fire extinguishing equipment through intelligent task allocation and dynamic path optimization. Path optimization ensures that the equipment can quickly reach the task point, and the real-time adjustment mechanism ensures that the equipment can still maintain efficient operation under the rapid changes of the fire environment. Under the combined effect, the module significantly improves the execution efficiency of fire extinguishing tasks, while reducing equipment operation risks and fire control costs.

[0073] Preferably, the task allocation unit of the adaptive fire extinguishing control module allocates tasks to drones and fire extinguishing robots based on fire dynamic prediction data, and sets the execution order according to the importance level of the task area to ensure maximum resource utilization and improved fire control efficiency.

[0074] The adaptive fire extinguishing control module in the present invention intelligently allocates fire extinguishing tasks to drones and fire extinguishing robots through a task allocation unit according to the fire dynamic prediction data. The task allocation unit adopts a multi-objective optimization algorithm to set the task execution order based on the importance level of the fire area to ensure the optimal use of fire extinguishing resources and improve the efficiency of fire control.

[0075] The task allocation unit first receives the fire dynamic prediction data generated by the fire modeling and prediction module, including the location of high-risk fire areas, the direction and speed of flame spread, and the dynamic changes of environmental variables (such as wind speed and humidity). After preprocessing, these data are used as input to the task allocation algorithm to analyze the urgency and regional priority of the fire.

[0076] According to the fire dynamic prediction data, the task allocation unit divides the task area into grades. The specific classification standards include: High priority areas: including areas with higher flame height, greater burning intensity and faster spreading speed; Medium priority area: an area with moderate flame height and burning intensity and stable diffusion trend; Low priority areas: areas where the fire is weak and the risk of spreading is low.

[0077] By calculating the potential threat of fire spread to surrounding equipment, tanks and personnel, the system assigns an importance score to each task area. The higher the score, the higher the priority.

[0078] Based on the importance level of the task area, the task allocation unit uses a multi-objective optimization algorithm (such as particle swarm optimization algorithm or ant colony algorithm) to allocate tasks. The optimization objectives include: Maximize the utilization rate of fire-fighting equipment; minimize the path cost of equipment reaching the task point; balance the task load of each device.

[0079] The core formula for optimization calculation is as follows: , in, represents the optimization objective function; Indicates The path cost of each device; Indicates the time required for the device to complete the task; Indicates the current task load of the device; , , Represents the weight coefficient, which is used to adjust the priority of different optimization objectives. The task allocation results in the UAV being assigned to the high-priority area for rapid fire extinguishing, the robot being responsible for flame control and local cooling in the medium-priority area, and the low-priority area being arranged in subsequent tasks.

[0080] The task allocation unit achieves optimal utilization of fire-fighting equipment resources by intelligently allocating fire-fighting tasks and setting a reasonable execution order. The system can quickly respond to dynamic changes in fires, prioritize high-risk areas, effectively curb the trend of fire spread, and improve the overall efficiency and accuracy of firefighting. At the same time, the multi-objective optimization algorithm ensures the reasonable scheduling of equipment, avoids resource waste and overload operation, and significantly improves the execution quality of fire-fighting tasks.

[0081] Example: A fire broke out in a petrochemical tank area. The fire spread rapidly and the tanks were close to each other, posing a major threat to neighboring tanks. Drones and firefighting robots were deployed on site, and these devices needed to be efficiently dispatched to control the fire and prevent the tanks from exploding. Implementation process: The system receives fire dynamic prediction data and identifies high-risk area A, medium-risk area B, and low-risk area C. High-risk area A is located near the storage tank, with a flame height of more than 5 meters, high burning intensity, and a diffusion direction directly pointing to the adjacent storage tank.

[0082] Based on the potential threat of the area to surrounding facilities, area A is assigned the highest priority, medium-risk area B is assigned a medium priority, and low-risk area C is assigned the lowest priority.

[0083] The system assigns two drones to high-risk area A to focus on firefighting, and two robots to medium-risk area B to perform combustion suppression and local cooling. The particle swarm optimization algorithm is used to calculate the shortest path for drones and robots to reach the task point, and set the execution order.

[0084] The drone prioritizes controlling the fire in area A, using high-pressure water mist for rapid cooling; the robot sprays foam in area B to suppress combustion. After the mission is completed, the equipment returns to the backup point according to the optimized path and stands by to respond to the next mission.

[0085] Through reasonable task allocation and execution, the system quickly contained the fire in the high-risk area at the early stage of the fire and prevented the fire from spreading to adjacent tanks. The fire in the medium-risk area was effectively controlled, and the fire in the low-risk area was completely extinguished in subsequent tasks. The entire fire-fighting process was efficient and accurate, the equipment utilization rate was maximized, and the impact of the fire on the tank area was minimized.

[0086] Preferably, Figure 4 As shown, the path optimization unit of the adaptive fire extinguishing control module generates the optimal moving path by using a path planning algorithm based on a cost function by updating the obstacle distribution in the fire area, the current position of the equipment and the shortest path cost in real time, thereby ensuring that the fire extinguishing equipment reaches the task point efficiently.

[0087] The path optimization unit of the adaptive fire extinguishing control module plays a vital role in the petroleum and petrochemical fire linkage control device. Its core function is to generate the optimal moving path of the fire extinguishing equipment based on the real-time updated obstacle distribution in the fire area, the current position of the equipment and the shortest path cost, using the path planning algorithm to ensure that the equipment can reach the task point quickly and efficiently to perform the fire extinguishing task.

[0088] The path optimization unit uses a path planning algorithm based on a cost function to transform the path planning problem into a cost minimization problem. During the path search process, the obstacles in the fire area, the direction of flame spread, the current location of the equipment, and environmental variables (such as wind speed and humidity) are comprehensively considered to dynamically calculate the path cost, thereby achieving dynamic adjustment of path planning.

[0089] The path optimization unit obtains real-time fire dynamic prediction data through the fire modeling and prediction module, including flame spread direction, risk area distribution and environmental variables. At the same time, it receives the current location information of the equipment and the obstacle distribution data in the fire area from the multimodal data acquisition module. After fusion processing, these data are used to calculate the basic input parameters required for the optimal path.

[0090] The path optimization unit uses a path planning method that combines the A* algorithm (A-Star Algorithm) with dynamic programming. The A* algorithm finds the shortest path from the current location of the device to the target task point through heuristic search, and uses the path cost function as the optimization target. The path cost function is defined as follows: , in, represents the total path cost; Indicates the distance cost of the device from the current node to the next node; It represents the obstacle risk cost, reflecting the risk level of obstacles in the path; Represents the cost of environmental variables, including the impact of factors such as wind speed and humidity on path selection; , , The weight coefficient representing the cost function is dynamically adjusted according to the fire scenario. The algorithm gradually generates the optimal moving path from the current position of the equipment to the target point by searching for the node with the minimum cost in the path.

[0091] The fire scene environment is complex and changeable, and the path optimization unit has the ability to adjust in real time. When the fire spreads or the position of obstacles changes, the path optimization unit recalculates the path cost based on the latest data and dynamically updates the optimal path to ensure that the equipment can avoid risks and quickly reach the target task point.

[0092] The path optimization unit generates a dynamically adjusted optimal mobile path by updating key data in the fire area in real time, ensuring efficient scheduling and accurate deployment of fire-fighting equipment. The path planning algorithm based on the cost function can effectively avoid obstacles and high-risk areas in complex fire environments, greatly improving the response speed and operating efficiency of fire-fighting equipment. In addition, the dynamic adjustment capability of the path optimization unit enables it to adapt to the rapid changes in the fire scene, thereby minimizing the risk of equipment operation and improving the overall fire control efficiency.

[0093] Example: In the emergency response to a sudden fire in a petrochemical tank area, the flames quickly spread to the narrow passages between the tanks, accompanied by a lot of smoke and high temperature. There are multiple obstacles around the tanks, including pipe supports and stored materials. Firefighting equipment needs to quickly reach high-risk areas to extinguish the fire, but the path at the fire scene is complex and the environment changes rapidly, so the accuracy of equipment scheduling is crucial. Implementation process: The multimodal data acquisition module collects real-time data at the fire scene, including obstacle distribution, equipment current location information, and fire dynamic prediction data. Data on flame spread direction, wind speed, and high-risk areas are input into the path optimization unit.

[0094] The path optimization unit uses the A* algorithm combined with the dynamic programming method based on the input data to calculate the shortest path from the current position of the drone to the high-risk area. During the path search process, the system dynamically evaluates the risk of obstacles in the path and the impact of environmental variables on the device's movement, and determines the optimal choice for each node in the path through the cost function.

[0095] As the drone approaches the high-risk area, the flame spreads faster and some nodes in the original path are covered by flames. The path optimization unit receives the latest prediction data from the fire modeling and prediction module, updates the path cost in real time, and replans the optimal path to avoid the flames.

[0096] The drone moved to the target mission point along the optimized path, using high-pressure water mist to quickly cool the flames and successfully control the fire. The robot simultaneously followed the optimized path to reach the medium-risk area and cooled the high-temperature area near the storage tank to prevent the fire from spreading.

[0097] Through real-time path planning and dynamic adjustment of the path optimization unit, the fire extinguishing equipment quickly reaches the task point at the early stage of the fire, effectively controlling the spread of the fire and preventing the fire from further spreading to adjacent tanks. The entire fire extinguishing process is efficient and accurate, the equipment operation time is shortened by 30%, the fire control efficiency is significantly improved, and the economic losses of the tank area are minimized.

[0098] like Figure 3 As shown, the virtual reproduction and optimization module is used to receive feedback data of the fire extinguishing process, generate a virtual scene for virtual reproduction of the fire scene, and use the virtual scene to train operators and optimize equipment configuration, and optimize the modeling parameters of the adversarial network and the path prediction parameters of the differential equation network based on the feedback data.

[0099] The virtual reproduction and optimization module is a key component of the petroleum and petrochemical fire linkage control device. It is designed to generate virtual fire scenes by receiving feedback data from the fire extinguishing process to support operator training and equipment configuration optimization. The module also dynamically optimizes the modeling parameters of the generative adversarial network (GAN) and the path prediction parameters of the differential equation network (ODE-Net) in the fire modeling and prediction module based on the feedback data, thereby improving the accuracy of fire modeling and prediction.

[0100] The virtual reproduction and optimization module uses the feedback data of the fire extinguishing process (including equipment operation status, fire extinguishing efficiency, and environmental parameter changes) to reconstruct the three-dimensional fire scene. The module uses holographic projection technology and virtual reality (VR) simulation technology to generate interactive virtual scenes containing key information such as flame dynamic changes, smoke diffusion, and equipment operation trajectory.

[0101] This virtual scene can not only truly restore the characteristics of the fire scene, but also dynamically display the trend of flame spread, providing operators with intuitive fire situation awareness. At the same time, operators can conduct operation drills in the virtual scene, including equipment scheduling, fire extinguishing path selection and emergency strategy adjustment.

[0102] The virtual reproduction and optimization module receives and analyzes the data fed back during the fire extinguishing process, including the error between the flame morphology and the predicted result, the deviation of the equipment operation, etc. These data are used to optimize the modeling parameters of GAN and the path prediction parameters of ODE-Net.

[0103] GAN modeling parameter optimization: By calculating the error between the virtual flame morphology data and the real flame morphology data, the network parameters of the generator and discriminator (such as learning rate and convolution kernel size) are adjusted to improve the accuracy of flame morphology modeling.

[0104] ODE-Net path prediction parameter optimization: According to the deviation between the virtual path and the actual equipment travel path, key parameters such as thermal diffusion coefficient and combustion reaction rate are dynamically adjusted to improve the accuracy of fire path prediction.

[0105] Virtual scenarios provide a low-risk, efficient way to train operators. Operators can repeatedly practice fire response strategies in a virtual environment, improving their decision-making and equipment operation skills.

[0106] At the same time, the equipment configuration optimization module based on virtual scenes can simulate the impact of different equipment layout schemes on the fire control effect, and provide the optimal configuration scheme for actual fire extinguishing tasks.

[0107] The virtual reproduction and optimization module provides an efficient training platform for operators by generating high-precision virtual fire scenes, significantly improving the efficiency and accuracy of fire-fighting tasks. The module's dynamic optimization of GAN and ODE-Net parameters further improves the accuracy of fire modeling and prediction, ensuring the rapid response and scientific decision-making of the fire linkage system in complex fire scenes. At the same time, equipment configuration optimization significantly reduces the waste of fire-fighting resources and provides strong technical support for the efficient control of petrochemical tank fires.

[0108] Preferably, the holographic scene reconstruction unit of the virtual reproduction and optimization module generates a dynamic scene including flame height, spreading speed and equipment position by performing three-dimensional reconstruction on multimodal data of the fire scene.

[0109] The holographic scene reconstruction unit in the virtual reproduction and optimization module generates a dynamic virtual scene including flame height, spread speed and real-time position of equipment by performing three-dimensional reconstruction of multimodal data of the fire scene (such as thermal data, smoke concentration, equipment location and environmental parameters). This unit uses three-dimensional point cloud data reconstruction technology and time series analysis methods to intuitively present the dynamic development process of the fire and the working status of the equipment, providing operators with real-time global situation awareness, and providing high-precision scene data support for fire modeling and prediction.

[0110] The holographic scene reconstruction unit first receives real-time data from the multimodal data acquisition module, including: The flame temperature distribution collected by the thermal imaging sensor provides information on flame height and heat intensity; the smoke particle concentration obtained by the laser scattering system reflects the flame diffusion path; the vibration signal detected by the piezoelectric sensor identifies the change in combustion intensity in the high-temperature area; the equipment location information recorded by the GPS module is used to reconstruct the distribution and movement trajectory of the equipment in the fire scene. After normalization and filtering, these data serve as the input basis for three-dimensional reconstruction.

[0111] Holographic scene reconstruction uses 3D point cloud reconstruction technology based on sparse convolutional network. The specific steps are as follows: (1) Map the multimodal data into three-dimensional space to generate a preliminary point cloud structure of the fire scene. Each point in the point cloud represents the height, temperature, or equipment location of the flame.

[0112] (2) Dynamically update the point cloud data through time series analysis, remove redundant information, and retain key flame boundaries, equipment trajectories, and high-temperature areas.

[0113] (3) The optimized point cloud is input into a 3D rendering engine and combined with the time series data of flame spread to generate a dynamic fire scene that can display the flame spread speed, direction, and height changes in real time.

[0114] The holographic scene provides interactive functions through virtual reality (VR) devices. Operators can interact with the virtual scene through gestures or voice, query the flame height, temperature or equipment status of a specific area in real time, and adjust equipment deployment and fire extinguishing strategies.

[0115] The holographic scene reconstruction unit dynamically restores the fire scene through 3D reconstruction technology, accurately displays the height of the flames, the speed of spread and the real-time location of the equipment, provides operators with clear situational awareness and improves command and decision-making capabilities. The real-time interactive function of the dynamic scene significantly improves the efficiency of equipment dispatch and the utilization rate of fire-fighting resources, while providing reliable data support for fire modeling and path prediction.

[0116] Example: A fire broke out in a petrochemical tank area. The flames spread rapidly and the smoke density was high. The visibility on site was poor. Traditional fire monitoring methods could not provide accurate fire dynamic information, and the fire fighting task faced challenges. To ensure the accurate deployment of fire fighting equipment, it is necessary to reconstruct the fire scene in real time to assist the operator in making decisions. Implementation process: Data collection and processing: The thermal imaging sensor collects the temperature distribution of the flame around the tank and generates a two-dimensional thermal temperature map. The laser scattering system monitors the smoke diffusion path and obtains smoke concentration data. The piezoelectric sensor records vibration signals and identifies changes in combustion intensity. The GPS module collects real-time location information of the drone and fire-fighting robot. The above data is normalized by the data processing unit, and after removing noise, it is transmitted to the holographic scene reconstruction unit.

[0117] Based on the multimodal data of the fire scene, a 3D point cloud is generated to preliminarily construct the spatial framework of the fire scene. Combined with the time series data of flame spread, the 3D point cloud is dynamically updated to optimize the display of flame boundaries, equipment trajectories and high-temperature areas. A dynamic visualization of the fire scene is generated through a 3D rendering engine, including flame height, spread speed and real-time location of equipment.

[0118] Operators wear VR equipment to observe fire scenes in real time and understand the flame dynamics and equipment distribution in high-risk areas. Operators use the virtual interface to adjust the spray angle of the drone and the position of the fire-fighting robot, and optimize equipment deployment according to the flame spread speed in the virtual scene.

[0119] The system feeds back data such as flame height and equipment trajectory in the virtual scene to the fire modeling module to optimize flame morphology modeling and path prediction parameters. According to the optimal fire extinguishing strategy simulated in the virtual scene, the execution plan of the actual fire extinguishing task is adjusted.

[0120] Through the holographic scene reconstruction unit, the system successfully generated a dynamic three-dimensional picture of the fire scene, providing operators with a clear understanding of the fire situation and optimizing equipment deployment and fire extinguishing paths. Precise control at the early stage of the fire significantly reduced the spread of the fire, and ultimately effectively controlled the fire in a short period of time, reducing the risk of tank explosion and economic losses.

[0121] Preferably, the virtual reproduction and optimization module superimposes the holographic scene into the operator's field of view through an augmented reality auxiliary system to display the real-time spread path of the fire, the operating status of the fire extinguishing equipment and changes in the surrounding environment.

[0122] The augmented reality (AR) auxiliary system in the virtual reproduction and optimization module can visualize the real-time spread path of the fire, the operating status of the fire extinguishing equipment and the changes in the surrounding environment by superimposing the fire holographic scene into the operator's field of view. Based on multimodal data fusion and real-time rendering technology, the system can dynamically display key information of the fire scene and provide an interactive command platform for operators, thereby optimizing the dispatch of fire extinguishing equipment and the execution of fire extinguishing strategies.

[0123] The AR auxiliary system receives multimodal data from the holographic scene reconstruction unit to generate a three-dimensional dynamic fire scene, which specifically includes: Fire spread path: The direction and speed of flame spread calculated by the Fire Modeling and Prediction Module; Equipment operation status: real-time location and operation parameters (such as moving speed and spray intensity) of equipment such as drones and fire-fighting robots; Ambient data: External conditions such as wind speed, humidity, and tank distribution information.

[0124] After real-time rendering and projection processing, these data are superimposed on the operator's field of view to form a dynamic and interactive AR scene.

[0125] The AR assistance system projects a three-dimensional fire scene into the operator's actual field of view by wearing an AR device (such as a smart helmet or AR glasses). The system uses spatial mapping and object recognition technology to ensure accurate alignment of virtual information with the real scene. Specifically: Spatial mapping, using depth sensors to obtain the spatial structure of the fire scene, constructing a mapping relationship between the virtual fire scene and the real space; information overlay, overlaying the virtual flame shape, equipment location and environmental parameters in real time to the corresponding real location in the form of annotations, dynamically updating key information such as flame height and diffusion path; interactive function, operators can select specific areas or equipment through gestures or voice commands to view detailed data, and adjust fire-fighting strategies based on real-time changes.

[0126] Through the augmented reality assistance system, operators can intuitively grasp the dynamic changes of the fire scene and fully understand the flame spread path, high-risk areas and equipment operation status. This visual display of information significantly improves the operator's situational awareness, speeds up decision-making, and improves the accuracy of fire-fighting tasks and the efficiency of equipment scheduling. The interactive function of the system further enhances the operator's control ability and can optimize the fire-fighting strategy in real time, thereby effectively curbing the spread of fire and reducing fire losses.

[0127] Preferably, the virtual reproduction and optimization module adjusts the generator learning rate and discriminator error function parameters of the generative adversarial network in real time by analyzing the feedback data of the fire extinguishing process, and optimizes the physical field parameter calculation formula in the differential equation network.

[0128] The virtual reproduction and optimization module analyzes the feedback data of the fire extinguishing process, adjusts the generator learning rate and discriminator error function parameters of the generative adversarial network (GAN) in real time, and optimizes the physical field parameter calculation formula in the differential equation network (ODE-Net), thereby improving the accuracy and real-time performance of flame morphology modeling and fire path prediction.

[0129] The module receives multimodal feedback data collected during the fire extinguishing process, including flame morphology, spread path, equipment operating status, and dynamic changes in environmental parameters (such as wind speed and humidity). After preprocessing, these data are compared with the predicted data of the fire modeling module to calculate the prediction error, which mainly includes: Flame morphology error: the morphological difference between the generated flame image and the real flame image. Path prediction error: the deviation between the predicted fire spread path and the actual observed path of the equipment.

[0130] The core of GAN consists of a generator and a discriminator. The generator is used to generate flame morphology data, and the discriminator is used to determine the authenticity of the generated data. According to the feedback data of the fire extinguishing process, the system dynamically adjusts the learning rate of the GAN generator and the error function parameters of the discriminator: Adjustment of the generator learning rate: According to the flame morphology error, the learning rate is optimized through the gradient descent method, so that the generator can gradually generate flame morphology data that is closer to the real fire scene during training.

[0131] Optimization of the discriminator error function parameters: By adjusting the weight coefficient in the error function, the recognition ability of key features such as flame edge and combustion intensity is enhanced, thereby improving the discriminator's discrimination accuracy.

[0132] ODE-Net models the fire path through the heat conduction equation and the combustion reaction rate equation. The system dynamically adjusts the physical field parameters in ODE-Net, including the thermal diffusion coefficient, combustion reaction rate, and external environmental disturbance terms, based on the path prediction error: , in, represents the thermal diffusion coefficient; represents the combustion reaction rate coefficient; represents the external environment disturbance coefficient; Represents the wind speed vector field. By optimizing these parameters, the accuracy of fire path prediction is improved so that it can better reflect the actual dynamics of fire spread.

[0133] Through dynamic optimization of GAN and ODE-Net, the virtual reproduction and optimization module significantly improves the accuracy and real-time performance of fire modeling and prediction. The optimized GAN can generate flame morphology data that highly matches the actual flame characteristics, providing accurate support for the dispatch of fire-fighting equipment; the optimized ODE-Net can accurately predict the fire path and provide a reliable basis for the formulation of fire-fighting strategies. At the same time, this real-time optimization mechanism enables the fire linkage control device to quickly respond to dynamic changes at the fire scene, greatly improving the execution efficiency of fire-fighting tasks.

[0134] Example: A fire broke out in a petrochemical tank area, and the flames quickly spread to the surrounding area. The height and spread path of the flames are constantly changing due to the influence of the ambient wind speed and the distribution of obstacles around the tanks. In order to ensure the accuracy of fire modeling and prediction, it is necessary to optimize the core parameters of flame morphology modeling and fire path prediction in real time. Implementation process: The module receives flame height, temperature distribution and path data fed back by the fire extinguishing equipment. The actual observation data is compared with the flame shape data generated by GAN to calculate the flame shape error. The equipment travel path is compared with the fire spread path predicted by ODE-Net to calculate the path prediction error.

[0135] According to the flame morphology error, the learning rate of the generator is adjusted to make the flame morphology generated by the generator more realistic. By adjusting the discriminator error function parameters, the recognition ability of the flame edge and high temperature area is enhanced, and the flame morphology modeling accuracy is optimized.

[0136] According to the path prediction error, the thermal diffusion coefficient and combustion reaction rate parameters of ODE-Net are optimized to make the predicted path closer to the actual path. The external environmental disturbance term is dynamically adjusted to consider the impact of wind speed changes on the fire spread path to ensure the real-time and accuracy of the prediction results.

[0137] The optimized flame shape data and fire path prediction data are output and transmitted to the fire extinguishing equipment control module. Based on the optimized data, the system replans the mission paths of drones and robots, adjusts the spray parameters, and improves the fire extinguishing efficiency.

[0138] The flame morphology data generated by the optimized GAN is closer to the actual fire scene, and the flame edge and combustion intensity characteristics are more accurately represented. The deviation between the predicted path of the optimized ODE-Net and the actual spread path is reduced by 20%, which greatly improves the dispatch accuracy of the fire extinguishing equipment. Through dynamic optimization, the spread of the fire is effectively controlled in the early stage, preventing the flame from spreading to the adjacent tank area, and minimizing economic losses and safety hazards.

[0139] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware.

[0140] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A fire linkage control device for petroleum and petrochemical industry, characterized in that: include: Multimodal data acquisition module, including multiple sensors for collecting thermal data, smoke particle data and vibration data; Used to fuse the collected data and generate initial fire scene data; The fire modeling and prediction module is used to receive the initial fire scene data, perform flame morphology modeling through the generative adversarial network, and perform path prediction through the differential equation network to generate fire dynamic prediction data; Adaptive fire extinguishing control module, which is used to allocate fire extinguishing tasks and optimize fire extinguishing paths according to fire dynamic prediction data. The optimized fire extinguishing path uses a path planning algorithm combined with a real-time adjustment mechanism to adjust the operating parameters of the fire extinguishing equipment; The virtual reproduction and optimization module is used to receive feedback data of the fire extinguishing process and generate a virtual scene for virtual reproduction of the fire scene. The virtual scene is used to train operators and optimize equipment configuration, and optimize the modeling parameters of the adversarial network and the path prediction parameters of the differential equation network based on the feedback data.

2. The fire linkage control device for petrochemical industry according to claim 1, characterized in that: The multimodal data acquisition module further includes an environmental humidity sensor and a wind speed sensor, and generates initial fire scene data by fusing the environmental humidity data and wind speed data with the thermal data, smoke particle data and vibration data.

3. The fire linkage control device for petrochemical industry according to claim 2, characterized in that: The mimetic neural network of the multimodal data acquisition module converts thermal data, smoke particle data, vibration data, environmental humidity data and wind speed data into a unified comprehensive feature vector by establishing an association model between time series features and spatial distribution features. The comprehensive feature vector is optimized by a graph Laplace dimensionality reduction algorithm to generate initial fire scene data.

4. The fire linkage control device for petrochemical industry according to claim 1, characterized in that: The generator of the generative adversarial network of the fire modeling and prediction module generates flame morphology data based on the input initial fire scene data, and the discriminator calculates the error between the generated flame morphology data and the real flame morphology data, and feeds the error back to the generator to adjust its network parameters until the generated flame morphology data matches the real flame morphology data.

5. The fire linkage control device for petrochemical industry according to claim 4, characterized in that: The differential equation network of the fire modeling and prediction module calculates the fire spread path by analyzing the physical field parameters of heat conduction characteristics, combustion reaction rate and fire spread direction, and dynamically adjusts the path prediction results to cope with real-time changes in environmental variables.

6. The fire linkage control device for petrochemical industry according to claim 1, characterized in that: The task allocation unit of the adaptive fire extinguishing control module allocates tasks to drones and fire extinguishing robots based on fire dynamic prediction data, and sets the execution order according to the importance level of the task area to ensure maximum resource utilization and improved fire control efficiency.

7. The fire linkage control device for petrochemical industry according to claim 6, characterized in that: The path optimization unit of the adaptive fire extinguishing control module generates the optimal moving path by updating the obstacle distribution in the fire area, the current position of the equipment and the shortest path cost in real time, and uses the path planning algorithm based on the cost function to ensure that the fire extinguishing equipment reaches the task point efficiently.

8. The fire linkage control device for petrochemical industry according to claim 1, characterized in that: The holographic scene reconstruction unit of the virtual reproduction and optimization module generates a dynamic scene including flame height, spreading speed and equipment position by performing three-dimensional reconstruction on the multimodal data of the fire scene.

9. The fire linkage control device for petrochemical industry according to claim 8, characterized in that: The virtual reproduction and optimization module superimposes the holographic scene into the operator's field of view through an augmented reality auxiliary system to display the real-time spread path of the fire, the operating status of the fire extinguishing equipment and changes in the surrounding environment.

10. The fire linkage control device for petrochemical industry according to claim 1, characterized in that: The virtual reproduction and optimization module analyzes the feedback data of the fire extinguishing process, adjusts the generator learning rate and discriminator error function parameters of the generative adversarial network in real time, and optimizes the physical field parameter calculation formula in the differential equation network.

Images