Fire control strategy acquisition method and device for wine cellar, computer device, medium and product

Abstract

The application relates to a wine cellar fire control strategy acquisition method and device, computer equipment, a medium and a product. The method comprises the following steps: acquiring multi-modal data in a wine cellar environment, compensating for drift of the multi-modal data to obtain target data, extracting visual spatial features of the target data through a target detection network model, extracting time sequence features of the target data through a machine learning network model, fusing the visual spatial features and the time sequence features to obtain fused features, performing fusion analysis on the fused features through a multi-modal analysis model to obtain fire risk grades and fire source positioning information, constructing a fire spread model according to the fire source positioning information and spatial structure information, acquiring a fire spread area according to the fire spread model, and acquiring a wine cellar fire control strategy according to the fire risk grades and the fire spread area. The method can improve detection accuracy.

Description

Technical Field

[0001] This application relates to the field of fire monitoring technology, and in particular to a method, device, computer equipment, medium and product for obtaining fire control strategies for wine cellars. Background Technology

[0002] Wine cellars, as core storage areas in the brewing industry, possess unique fire risk characteristics. Alcoholic substances are volatile, forming flammable gases (such as ethanol vapor), which can ignite upon contact with an ignition source, potentially causing a flowing fire or explosion. Dust generated during the processing of brewing raw materials (such as grain dust) poses a dust explosion risk when it reaches a certain concentration in the air. Furthermore, the high temperature and humidity environment in wine cellars accelerates the aging of electrical equipment, leading to frequent false alarms and missed alarms in traditional fire protection systems. These factors make wine cellars a key area for fire safety prevention, but traditional fire protection methods are insufficient to effectively address the complex needs of such environments.

[0003] Currently, fire monitoring in wine cellars mainly relies on single sensor technologies (such as smoke detectors, heat detectors, or combustible gas detectors) or simple combination alarm systems. These systems often use fixed threshold triggering mechanisms and cannot dynamically adapt to the risk differences in different areas within the wine cellar. Summary of the Invention

[0004] Therefore, it is necessary to provide a method, device, computer equipment, medium, and product for acquiring fire control strategies for wine cellars that can improve detection accuracy, in order to address the aforementioned technical problems.

[0005] Firstly, this application provides a method for obtaining a fire control strategy for a wine cellar, including:

[0006] Multimodal data in the wine cellar environment is acquired, and drift compensation is performed on the multimodal data to obtain the target data; the multimodal data includes visible light video data, infrared thermal imaging data, environmental sensor data, and combustible gas concentration data;

[0007] Visual spatial features of target data are extracted using an object detection network model, and temporal features of target data are extracted using a machine learning network model.

[0008] Visual spatial features and temporal features are fused to obtain fused features, and the fused features are analyzed by a multimodal analysis model to obtain fire risk level and fire source location information;

[0009] Based on the fire source location information and spatial structure information, a fire spread model is constructed, and the fire spread area is obtained based on the fire spread model;

[0010] Based on the fire risk level and the area of ​​fire spread, obtain the fire control strategy for the wine cellar.

[0011] In one embodiment, the step of performing drift compensation on multimodal data to obtain target data includes:

[0012] Based on the spatial structure information of the wine cellar, multiple grid areas were obtained, and environmental parameters for each grid area were acquired. The environmental parameters included baseline temperature and humidity and baseline dust level.

[0013] Constructing a digital twin model of the wine cellar;

[0014] Multimodal data and environmental parameters are mapped to a digital twin model, and the simulation results of the digital twin model are obtained;

[0015] Drift compensation is performed on multimodal data based on simulation results.

[0016] In one embodiment, the step of performing drift compensation on multimodal data based on simulation results includes:

[0017] The temperature and humidity data in the simulation results and multimodal data are fused and filtered to obtain temperature and humidity compensation values;

[0018] When dust interference is determined to exist in the grid area based on visible light video data, the multimodal data is attenuated based on the baseline dust level to obtain the dust compensation value.

[0019] In one embodiment, the fire source location information includes the flame area and the temperature anomaly area; the step of constructing a fire spread model based on the fire source location information and spatial structure information includes:

[0020] Obtain the motion vectors corresponding to the flame area and the temperature anomaly area, and obtain the flame motion trajectory based on visible light video data;

[0021] Principal component analysis is performed on all motion vectors to obtain the flame spread direction, and the flame spread velocity is obtained based on the flame trajectory.

[0022] The digital twin model is discretized into multiple regular grids on a preset plane, and the state attributes corresponding to the regular grids are determined; the state attributes include unburned, burning, burned out, and non-flammable.

[0023] Based on the direction and speed of flame spread, the ignition energy corresponding to the regular grid is obtained, and the grid ignition rules are configured for the regular grid according to the ignition energy and state attributes to obtain the fire spread model.

[0024] In one embodiment, the step of obtaining the fire spread area based on the fire spread model includes:

[0025] Using the direction and speed of flame spread as model driving signals, the fire spread model is solved iteratively in multiple rounds to obtain the fire spread area corresponding to multiple time periods.

[0026] In one embodiment, the step of obtaining a fire control strategy for the wine cellar based on the fire risk level and the area of ​​fire spread includes:

[0027] An alarm message is triggered when the fire risk level is Level 1.

[0028] When the fire risk level is Level 2, the target fire extinguishing equipment is determined based on the area of ​​fire spread, and the target fire extinguishing equipment is activated.

[0029] When the fire risk level is Level 3, the fire-fighting zone is determined based on the area of ​​fire spread, and all fire-fighting equipment corresponding to the fire-fighting zone is activated.

[0030] Secondly, this application also provides a wine cellar fire control strategy acquisition device, comprising:

[0031] The data compensation module is used to acquire multimodal data in the wine cellar environment and perform drift compensation on the multimodal data to obtain the target data; the multimodal data includes visible light video data, infrared thermal imaging data, environmental sensor data, and combustible gas concentration data;

[0032] The feature extraction module is used to extract the visual spatial features of the target data through the target detection network model and the temporal features of the target data through the machine learning network model.

[0033] The feature fusion module is used to fuse visual spatial features and temporal features to obtain fused features, and then to perform fusion analysis on the fused features through a multimodal analysis model to obtain fire risk level and fire source location information;

[0034] The fire prediction module is used to construct a fire spread model based on fire source location information and spatial structure information, and to obtain the fire spread area based on the fire spread model.

[0035] The strategy acquisition module is used to acquire fire control strategies for the wine cellar based on the fire risk level and the area of ​​fire spread.

[0036] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method steps of any one of the first aspects.

[0037] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method steps of any one of the first aspects.

[0038] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the method steps of any one of the first aspects.

[0039] The aforementioned method, device, computer equipment, medium, and product for acquiring fire control strategies in wine cellars acquire multimodal data in the wine cellar environment and perform drift compensation on the multimodal data to obtain target data. A target detection network model extracts the visual spatial features of the target data, and a machine learning network model extracts the temporal features of the target data. The visual spatial features and temporal features are fused to obtain fused features, which are then analyzed using a multimodal analysis model to obtain fire risk level and fire source location information. Based on the fire source location information and spatial structure information, a fire spread model is constructed. The fire spread area is obtained based on the fire spread model, and the fire control strategy for the wine cellar is obtained based on the fire risk level and the fire spread area. This reduces the probability of false alarms, improves the accuracy of fire risk level detection, and achieves precise fire source location and accurate fire prediction, thereby improving the reliability of the fire control strategy. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 This is an application environment diagram of the wine cellar fire control strategy acquisition method in one embodiment;

[0042] Figure 2 This is a flowchart illustrating a method for obtaining fire control strategies for a wine cellar in one embodiment;

[0043] Figure 3 This is a flowchart illustrating the method for obtaining the fire control strategy for a wine cellar in another embodiment;

[0044] Figure 4 This is a structural block diagram of a wine cellar fire control strategy acquisition device in one embodiment;

[0045] Figure 5 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0047] The wine cellar fire control strategy acquisition method provided in this application embodiment can be applied to, for example... Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104 or placed on a cloud or other network server. Terminal 102 is used to acquire multimodal data in the wine cellar environment, perform drift compensation on the multimodal data to obtain target data, extract the visual spatial features of the target data through a target detection network model, and extract the temporal features of the target data through a machine learning network model. The visual spatial features and temporal features are fused to obtain fused features, and the fused features are analyzed through a multimodal analysis model to obtain the fire risk level and fire source location information. Based on the fire source location information and spatial structure information, a fire spread model is constructed. The fire spread area is obtained based on the fire spread model, and the wine cellar fire control strategy is obtained based on the fire risk level and the fire spread area. Terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, drones, low-altitude aircraft, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, and projection equipment. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted displays. Head-mounted displays can be virtual reality (VR) devices, augmented reality (AR) devices, and smart glasses. Server 104 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0048] In one exemplary embodiment, such as Figure 2 As shown, a method for obtaining fire control strategies for wine cellars is provided, which can be applied to... Figure 1 Taking terminal 102 as an example, the explanation includes the following steps 202 to 210. Wherein:

[0049] S202: Acquire multimodal data in the wine cellar environment and perform drift compensation on the multimodal data to obtain target data; the multimodal data includes visible light video data, infrared thermal imaging data, environmental sensor data, and combustible gas concentration data.

[0050] Optionally, to achieve comprehensive sensing, a multimodal sensor array needs to be deployed in key areas of the wine cellar, such as the alcohol storage area, aging area, and workshop. This array includes high-definition infrared PTZ cameras, infrared thermal imagers, explosion-proof ethanol concentration sensors, temperature and humidity composite sensors, and laser dust sensors. Considering the characteristics of the wine cellar, the combustible gas concentration sensor needs to be optimized for ethanol vapor, and it should be installed 30-60 cm above the ground in the alcohol storage area to effectively monitor volatile gases. Before fusion analysis, the collected raw data needs to undergo data preprocessing and compensation, including spatiotemporal alignment and normalization of multi-source heterogeneous data, and environmental compensation algorithms to correct sensor data drift in response to common wine cellar conditions such as high humidity, high temperature, and dust interference, thereby reducing the false alarm rate. Visible light video data is used to capture visual appearances such as the shape, color, and diffusion outline of flames / smoke; infrared thermal imaging data is used for non-contact identification of abnormal high-temperature points, temperature distribution, and the core area of ​​flames, unaffected by smoke / strong light obstruction; environmental sensor data includes environmental time-series parameters such as temperature, humidity, dust concentration, and smoke concentration, reflecting dynamic changes in the environment; and combustible gas concentration data is used to monitor the concentration of ethanol vapor in wine cellars and detect early signs of fire such as alcohol leaks.

[0051] Furthermore, considering the special environment of the wine cellar with high humidity, high temperature and high dust, a partitioned adaptive compensation algorithm based on digital twins is used to correct the original multimodal data, and finally output target data that removes environmental interference and closely resembles the real environment, providing a high-quality data foundation for subsequent feature extraction.

[0052] S204: Extract the visual spatial features of the target data through the target detection network model, and extract the temporal features of the target data through the machine learning network model.

[0053] Optionally, visual spatial features refer to using an improved YOLO target detection network (such as YOLOv10) to extract features from visible light video and infrared thermal imaging data. The core principle is to use a lightweight convolutional neural network (CNN) to extract spatial dimension features such as flame morphology, flicker frequency, high-temperature area contours, and smoke texture and diffusion direction from image / video frames, thereby achieving a quantitative representation of the visual appearance of a fire. Temporal features refer to using machine learning network models (such as Long Short-Term Memory networks, LSTM) to extract features from temporal data such as environmental sensor data and combustible gas concentration data. The core principle is to use the gating mechanism of a recurrent neural network to capture the changing trends, fluctuation cycles, and abnormal mutation patterns (such as a sudden increase in ethanol concentration and a continuous rise in temperature) of temperature, humidity, dust, combustible gas, and smoke concentration, solving the problem that ordinary neural networks cannot process temporal data, and achieving a quantitative representation of the dynamic development of a fire.

[0054] S206: The visual spatial features and temporal features are fused to obtain fused features, and the fused features are analyzed by a multimodal analysis model to obtain the fire risk level and fire source location information.

[0055] Optionally, a fusion method driven by correlation and attention gating is employed to weightedly fuse visual spatial features and temporal features. By calculating the correlation matrix of the two types of features (e.g., the degree of synergy between high-temperature areas and flame pixels, the matching relationship between the smoke concentration rise rate and the thermal imaging temperature gradient), the correlation is used as an attention weight to dynamically control the fusion ratio of different features, generating a deep fusion feature vector that can cross-validate multimodal evidence. This retains the core information of both visual and temporal aspects while eliminating redundant and interfering features. Subsequently, the deep fusion feature vector is input into a trained multimodal artificial intelligence analysis model. The model outputs results through dual fusion analysis at the feature level and decision level. Feature-level fusion refers to in-depth feature mining of the fused features, strengthening fire-related features and weakening environmental interference features. Decision-level fusion refers to mapping the fused features to a probability distribution of multiple fire risk levels using a fully connected neural network and a Softmax function, selecting the highest probability as the final risk level. Simultaneously, through the model's localization branch, combined with the spatial coordinates of visual features and the location of high-temperature points in infrared thermal imaging, precise fire source localization is achieved.

[0056] S208: Construct a fire spread model based on fire source location information and spatial structure information, and obtain the fire spread area based on the fire spread model.

[0057] Optionally, based on the fire source location information, the motion vector of the flame between consecutive frames is analyzed using the optical flow method to predict the direction and speed of flame spread. Combined with the spatial structure information of the wine cellar, such as the shelf layout and aisles, a fire spread model is constructed to provide a dynamic adjustment basis for fire linkage control strategies.

[0058] S210: Obtain fire control strategies for wine cellars based on fire risk levels and fire spread areas.

[0059] Optionally, based on the fire risk level and fire source location information, corresponding fire-fighting linkage control strategies are generated and executed, including activating intelligent water guns for targeted fire extinguishing, controlling the ventilation system, and sending alarm information to terminals. For example, when the fire risk level is Level 1, a local audible and visual alarm is triggered and management personnel are notified to conduct on-site verification. When the fire risk level is the more severe Level 2, the intelligent water gun closest to the fire source is automatically activated for flame-retardant spraying, ensuring that the coverage area of ​​the fire extinguishing equipment is accurately matched with the spread area, avoiding waste of fire extinguishing resources and secondary damage.

[0060] The aforementioned method for acquiring fire control strategies for wine cellars involves obtaining multimodal data from the wine cellar environment and performing drift compensation on the multimodal data to obtain target data. A target detection network model is used to extract the visual spatial features of the target data, and a machine learning network model is used to extract the temporal features of the target data. The visual spatial features and temporal features are then fused to obtain fused features. A multimodal analysis model is used to perform fusion analysis on the fused features to obtain fire risk levels and fire source location information. Based on the fire source location information and spatial structure information, a fire spread model is constructed. The fire spread area is obtained based on the fire spread model. Finally, based on the fire risk level and the fire spread area, a fire control strategy for the wine cellar is acquired. This method can reduce the probability of false alarms, improve the accuracy of fire risk level detection, and achieve precise fire source location and accurate fire prediction, thereby improving the reliability of the fire control strategy.

[0061] In an exemplary embodiment, the step of performing drift compensation on multimodal data to obtain target data includes: obtaining multiple grid regions based on the spatial structure information of the wine cellar, and obtaining environmental parameters for each grid region; the environmental parameters include baseline temperature and humidity and baseline dust level; constructing a digital twin model of the wine cellar; mapping the multimodal data and environmental parameters to the digital twin model, and obtaining the simulation results of the digital twin model; and performing drift compensation on the multimodal data based on the simulation results.

[0062] Optionally, based on the spatial structure information of the wine cellar (such as geometric layout, functional zoning, and storage location), the wine cellar is divided into multiple independent grid areas (such as ceramic jar storage area, bottled wine area, blending workshop, and ventilation dead zone area). The environmental characteristics, operating scenarios, and interference types of each grid area are consistent. For each grid area, environmental parameter benchmarks under normal operating conditions are collected and calibrated. The core parameters are baseline temperature and humidity (the normal fluctuation range of temperature and humidity in this area when there is no interference or anomaly) and baseline dust level (the baseline value of dust concentration in this area during daily operation and when there is no fire). These parameters are the core reference for determining whether the data has drifted due to environmental interference. Based on the spatial structure information of the wine cellar, a 3D digital twin model is built, integrating the physical features of the wine cellar, such as geometric structure (three-dimensional dimensions / locations of walls, columns, and passages), functional zoning (corresponding one-to-one with the grid areas), ventilation layout, and storage distribution. By precisely linking environmental parameters such as baseline temperature and humidity and baseline dust levels of each grid area to the corresponding grid location in the digital twin model, the digital twin model not only replicates the physical space of the wine cellar but also incorporates the environmental baseline characteristics of each area, becoming a digital model that combines spatial and environmental attributes.

[0063] Furthermore, the raw data collected by the multimodal sensor array in the wine cellar (visible light video, infrared thermal imaging, temperature and humidity, dust, combustible gas concentration, etc.) is precisely mapped to the corresponding grid areas of the digital twin model according to the actual installation location of the sensors, achieving a one-to-one correspondence between physical data and digital space. Simultaneously, the environmental parameter benchmarks for each grid area are spatiotemporally aligned with the mapped raw data, ensuring that the raw data and environmental benchmarks can be directly compared and analyzed within the same grid area and at the same time stamp, eliminating misalignment interference caused by different acquisition locations and times for multi-source data. The digital twin model, based on the integrated spatial structure and environmental parameter benchmarks, combined with the environmental physical laws of the wine cellar (such as ethanol vapor diffusion, temperature and humidity conduction, and dust drift patterns), simulates the environmental physical field. During the simulation, the model calculates the expected real-time environmental parameter values ​​for each grid area based on the current environmental state (such as no fire, no abnormal operations). These expected values ​​are theoretically true values ​​after eliminating sensor drift and environmental interference, providing a core basis for subsequent compensation.

[0064] In this embodiment, multiple grid areas are obtained based on the spatial structure information of the wine cellar, environmental parameters of each grid area are obtained, a digital twin model of the wine cellar is constructed, multimodal data and environmental parameters are mapped to the digital twin model, and the simulation results of the digital twin model are obtained. Based on the simulation results, drift compensation is performed on the multimodal data, which can reduce sensor data drift, improve data authenticity, thereby effectively reducing the probability of false alarms and improving the accuracy of fire risk level detection.

[0065] In an exemplary embodiment, the step of performing drift compensation on multimodal data based on simulation results includes: performing fusion filtering on temperature and humidity data in the simulation results and multimodal data to obtain temperature and humidity compensation values; and, if dust interference is determined to exist in the grid area based on visible light video data, performing attenuation processing on the multimodal data based on the baseline dust level to obtain dust compensation values.

[0066] Optionally, the persistent high humidity and temperature environment of the wine cellar can cause hardware drift and random measurement errors in the temperature and humidity sensors (such as condensation on the sensor probe leading to higher values, or minor fluctuations in the circuit due to temperature), rather than actual environmental changes in the data itself. Therefore, a fusion filtering method is used to combine simulation results and measured data, and an algorithm is employed to estimate the temperature and humidity values ​​closest to the actual values. The simulation results of the digital twin model and the measured temperature and humidity values ​​from the multimodal data are used as the dual inputs to the filtering algorithm. Kalman filtering or particle filtering algorithms are used to fuse the dual input data. The core logic is to correct the error of the measured values ​​using the model's predicted values, while simultaneously updating the model's prediction accuracy using the measured values. The filtering algorithm continuously predicts, corrects, and updates the measured values, eliminating drift errors and random noise in the measured values, and outputs a temperature and humidity compensation value. This value is the optimal estimate of the actual temperature and humidity in this area of ​​the wine cellar, replacing the original measured values ​​for subsequent analysis.

[0067] Optionally, dust interference in wine cellars is mostly operational (such as dust generated from the detachment of seals on earthenware jars, handling of grain raw materials, and cleaning of wine jars), which, along with smoke particles in the early stages of a fire, are easily misjudged by dust sensors. Using visible light video data from the same grid area as the core discrimination criterion, combined with operational scene information from a digital twin model, we distinguish between interfering dust and fire-related smoke: if no open flame, smoke outline, or diffusion characteristics are detected in the video, but the measured value of the dust sensor rises sharply, and the numerical change pattern conforms to the dust diffusion model of handling, cleaning, and other operations in that area (such as diffusion along the passageway, a rapid increase followed by a decrease), it is determined to be interfering dust; if open flame / smoke characteristics are detected in the video, or the dust value change has no obvious operational correlation and shows a continuous upward trend, it is determined to be fire-related smoke, and no attenuation compensation is performed, retaining the original data. For the original data determined to be interfering dust, amplitude attenuation processing is performed based on the baseline dust level of that grid area. Specifically, if the measured dust value exceeds the baseline level but by a small margin, a slight attenuation is applied to correct the value to within the reasonable fluctuation range of the baseline. If the measured dust value significantly exceeds the baseline level (e.g., a sudden increase due to handling operations), a significant attenuation is applied to substantially reduce the weight of this data in subsequent fire risk analysis, preventing it from being misidentified as fire smoke. After attenuation processing, a dust compensation value is obtained, which eliminates invalid dust signals caused by operational interference and retains only dust data characteristics related to the fire situation.

[0068] In this embodiment, temperature and humidity compensation values ​​are obtained by fusing and filtering the temperature and humidity data in the simulation results and multimodal data. When dust interference is determined to exist in the grid area based on visible light video data, the multimodal data is attenuated based on the baseline dust level to obtain dust compensation values. This can accurately eliminate sensor drift and environmental interference, improve data authenticity, effectively reduce the probability of false alarms, and improve the accuracy of fire risk level detection.

[0069] In an exemplary embodiment, the fire source location information includes a flame region and a temperature anomaly region. The step of constructing a fire spread model based on the fire source location information and spatial structure information includes: obtaining motion vectors corresponding to the flame region and the temperature anomaly region, and obtaining the flame trajectory based on visible light video data; performing principal component analysis on all motion vectors to obtain the flame spread direction, and obtaining the flame spread speed based on the flame trajectory; discretizing the digital twin model into multiple regular grids on a preset plane, and determining the state attributes corresponding to the regular grids; the state attributes include unburned, burning, burnt out, and non-combustible; obtaining the ignition energy corresponding to the regular grids based on the flame spread direction and flame spread speed, and configuring grid ignition rules for the regular grids based on the ignition energy and state attributes to obtain the fire spread model.

[0070] Optionally, the microscopic motion features of the fire source are extracted from visual data using optical flow methods, while the macroscopic motion trajectory of the fire source is tracked using visible light video, providing raw data for subsequent calculations of the spread direction and velocity. For the flame region and temperature anomaly region in the fire source location information, their dense optical flow fields are calculated, transforming the pixel motion of the flame / high-temperature region into a two-dimensional pixel displacement vector (e.g., ...). , Each vector is associated with the temperature value of the infrared thermal image, accurately representing the movement direction and distance of each pixel within the fire source area, reflecting the microscopic diffusion characteristics of the fire. Simultaneously, cross-frame tracking is performed on the feature points with the largest brightness / temperature gradient of the flame front in the visible light video, recording the positional changes of these feature points in consecutive video frames to form the macroscopic movement trajectory of the flame. This trajectory is the core basis for subsequent calculations of the fire spread rate.

[0071] Furthermore, principal component analysis (PCA) is performed on all extracted motion vectors, and a weighted average direction, weighted by vector magnitude, is used for double verification. PCA extracts the most representative principal component direction (i.e., the main trend of fire spread) from the motion vectors, while the weighted average direction takes into account the intensity differences of motion between different pixels. The combination of the two eliminates the interference of random motion of local pixels, accurately determining the main spread direction of the flame. In addition, based on the positional changes of feature points in the flame trajectory, the instantaneous spread velocity of the feature points is calculated by the displacement distance of the feature points and the video frame interval. The average or maximum value of the velocities of multiple feature points is then taken as the overall spread velocity of the current fire. This process can be updated multiple times per second to ensure the real-time accuracy of the velocity.

[0072] Optionally, the 3D digital twin model constructed based on the spatial structure information of the wine cellar is discretized into multiple regular grids on a preset plane (usually the horizontal plane of the wine cellar, reflecting the main trend of horizontal fire spread). Each grid corresponds to a small area of ​​the physical space of the wine cellar, realizing the transformation of continuous space into discrete computing units, making spatial calculation of fire spread possible. Four mutually exclusive state attributes are set for each discretized regular grid to accurately represent the fire development state of the grid area, and the attributes can be dynamically updated as the model calculates. Among them, the unburned state indicates that there are combustibles in the grid but they have not yet been ignited, which is the initial state; the burning state indicates that the combustibles in the grid are burning and are the core unit for transferring ignition energy to the surrounding area; the burned-out state indicates that the combustibles in the grid are completely burned, there is no energy to transfer, and it cannot continue to spread; the non-combustible state indicates that there are no combustibles in the grid (such as columns, steel storage tanks, and cement floors in passageways), and the fire cannot spread in this area.

[0073] Furthermore, ignition energy is the core criterion for determining whether an unburned grid is ignited. Its calculation is driven by the direction and speed of flame spread, combined with the spatial structure information of the wine cellar (ventilation, distribution of combustibles), and weighted according to the physical laws of fire energy transfer. The core includes three parts: (1) Convective heat transfer energy: estimated based on the temperature of the fire source, the wind speed in the direction of spread (ventilation system status / preset value), and the distance between the grid and the fire source. The higher the wind speed and the closer the distance, the higher the convective heat transfer energy; (2) Radiative heat transfer energy: calculated based on the temperature of the grid during combustion and whether the line of sight between grids is blocked (such as pillars or shelves). If there is no obstruction and the temperature is higher, the radiative heat transfer energy is higher; (3) Fuel characteristic weighting: assigning a combustibility level weight to each grid (such as 1.0 for wooden shelves, 0.5 for the pottery jar area, and 0 for the non-combustible area). The higher the combustibility, the lower the energy threshold required for the grid to be ignited. Finally, the energy transfer efficiency is adjusted by combining the flame spread speed (the faster the speed, the wider the energy transfer range) to obtain the actual ignition energy and ignition energy threshold of each grid. Based on ignition energy and grid state attributes, a fire spread logic is established between grids, ultimately forming a wine cellar-specific fire spread model that can dynamically predict fire development. The core rules are: only burning grids can transfer ignition energy to surrounding unburned grids; when the total ignition energy received by an unburned grid is greater than or equal to its own ignition energy threshold, the grid's state attribute is updated from unburned to burning; burned / non-burnable grids do not participate in energy transfer or state updates, becoming the boundaries of fire spread.

[0074] In this embodiment, by acquiring the motion vectors corresponding to the flame area and the temperature anomaly area, and obtaining the flame trajectory based on visible light video data, principal component analysis is performed on all motion vectors to obtain the flame spread direction, and the flame spread speed is obtained based on the flame trajectory. The digital twin model is discretized into multiple regular grids on a preset plane, and the state attributes corresponding to the regular grids are determined. Based on the flame spread direction and flame spread speed, the ignition energy corresponding to the regular grids is obtained, and grid ignition rules are configured for the regular grids based on the ignition energy and state attributes to obtain the fire spread model. This model can achieve real-time dynamic prediction of fire spread, accurately capture the fire development trend, and thus improve prediction accuracy.

[0075] In an exemplary embodiment, obtaining the fire spread area based on the fire spread model includes: using the flame spread direction and flame spread speed as model driving signals, performing multiple rounds of iterative solutions on the fire spread model to obtain the fire spread area corresponding to multiple time periods.

[0076] Optionally, the model is iterated multiple times using the flame spread direction and velocity as driving signals. Essentially, this simulates the continuous spatial spread of the fire using discrete time steps. Given the rapid spread and development of fires in wine cellars, a very small iteration time step (e.g., 0.5 seconds / 1 second) is set to ensure the accuracy of the simulation results. The shorter the step, the closer it is to the real-time development of the fire. Within each time step, based on the current flame spread direction and velocity, the ignition energy of all burning grid areas to their surroundings is calculated. A state update of the entire cellar grid area is completed according to the state update rules, obtaining the fire spread state at that time point. The fire spread state of the previous iteration is used as the initial state for the next iteration. The constantly updated flame spread direction and velocity are continuously input, repeating the single-round iteration process to achieve continuous solution of the model. During the iteration process, the iteration results of the model are extracted at preset time nodes (e.g., 30 seconds, 60 seconds, 90 seconds). All grid areas in the burning state at that time node are spatially integrated to obtain the fire spread area for the corresponding time period. The entire iterative process is constrained by the spatial structure information of the wine cellar's digital twin model, including the wine cellar's geometry, storage layout, ventilation system, and fire-fighting facility locations, ensuring that the iterative results do not deviate from the actual physical space of the wine cellar, and giving the predicted spread area practical spatial reference value.

[0077] In this embodiment, by using the direction and speed of flame spread as model driving signals, the fire spread model is solved iteratively in multiple rounds to obtain the fire spread area corresponding to multiple time periods. This enables dynamic prediction in multiple time periods, accurately captures the spatiotemporal development trend of the fire, and thus improves the prediction accuracy.

[0078] In an exemplary embodiment, a fire control strategy for a wine cellar is obtained based on the fire risk level and the fire spread area, including: triggering an alarm message when the fire risk level is Level 1; determining the target fire extinguishing equipment based on the fire spread area when the fire risk level is Level 2, and controlling the activation of the target fire extinguishing equipment; and determining the enclosed fire extinguishing area based on the fire spread area when the fire risk level is Level 3, and controlling the activation of all fire extinguishing equipment corresponding to the enclosed fire extinguishing area.

[0079] Optionally, the first level is primary, characterized by no obvious open flame, only abnormal environmental parameters (such as a slight exceedance of ethanol vapor concentration or localized slight high temperature), or the fire is in its nascent stage with no risk of large-scale spread; the second level is intermediate, characterized by the presence of clear open flame / smoke, the precise location of the fire source, and the fire beginning to spread slowly, but not yet spreading to surrounding flammable areas (such as the ceramic jar storage area or alcohol storage tank area); the third level is advanced, characterized by an expansion of the open flame area, the fire spreading rapidly according to the predicted trend, threatening the core flammable area of ​​the wine cellar, and posing a risk of secondary disasters such as flowing fire and explosion.

[0080] Furthermore, when the fire risk level is determined to be Level 1, the fire is in its nascent or suspected stage, and there is no need to immediately activate fire extinguishing equipment. The core function is to achieve manual verification through alarm information, avoiding secondary losses such as water damage and equipment wear caused by accidental equipment activation. At this time, a local audible and visual alarm is automatically triggered (e.g., the alarm in the corresponding abnormal area of ​​the wine cellar), and at the same time, precise alarm information is pushed to the terminals of managers and safety officers (mobile phones, duty room screens), including the location of the abnormal area, the environmental parameters that triggered the alarm (such as ethanol concentration, temperature value), risk level, etc. The cause of the abnormality is confirmed only through manual on-site verification, distinguishing whether it is a real nascent fire or an abnormal parameter caused by environmental interference.

[0081] When the fire risk level is determined to be Level II, the fire has formed a clear source and has begun to spread locally. The core strategy is to pinpoint the fire extinguishing area by identifying the area of ​​fire spread, and then match the nearest / optimal target fire extinguishing equipment to achieve targeted flame retardation / extinguishing, preventing further fire spread. At this time, the fire spread area predicted by the fire spread model for the current and short periods (e.g., 30 seconds) is retrieved to clarify the core location of the fire source, the main direction of fire spread, and the local areas that will be affected. Combined with the precise spatial coordinates of fire-fighting facilities (smart water guns, fire hydrants, etc.) in the digital twin model of the wine cellar, the fire extinguishing equipment closest to the fire source and covering the core area of ​​fire spread is selected as the target fire extinguishing equipment. Only the target fire extinguishing equipment is activated to achieve directional spraying and targeted fire extinguishing. At the same time, the local ventilation system can be linked to prevent the accumulation of flammable gas from exacerbating the fire. Fire-fighting equipment in other areas remains on standby to reduce resource waste.

[0082] When the fire risk level is determined to be Level 3, the fire has entered a rapid development stage and poses a risk of widespread spread. The core strategy is to delineate an encirclement area based on the fire spread zone, and then activate all fire-fighting equipment within the area to achieve full-area fire suppression. Simultaneously, corresponding emergency measures are triggered to control the fire to the greatest extent possible. At this time, based on the fire spread model's prediction of the medium-to-long-term fire spread area (e.g., 60 / 90 seconds), combined with the wine cellar's spatial structure and combustible material distribution, an encirclement fire suppression zone is delineated. This zone includes the core fire source area, the entire affected area along the main fire spread direction, and surrounding critical flammable areas (e.g., the ceramic jar area and alcohol storage tank area). All intelligent water cannons within the encirclement fire suppression zone are activated, forming a multi-angle, full-coverage encirclement fire suppression posture, achieving dual control of flame retardation and fire suppression against the direction of fire spread. Simultaneously, non-fire-fighting power supplies in and around the encirclement area are cut off (to prevent electrical sparks from escalating the fire), the emergency smoke extraction system is activated (to remove smoke and combustible gases), the area's fireproof roller shutters are closed (to form physical fireproof isolation), and alarm information is sent to fire departments, achieving automatic system response and coordination with external rescue.

[0083] In this embodiment, by triggering an alarm when the fire risk level is Level 1, determining the target fire extinguishing equipment based on the fire spread area when the fire risk level is Level 2, and controlling the activation of the target fire extinguishing equipment when the fire risk level is Level 3, determining the surrounding fire extinguishing area based on the fire spread area, and controlling the activation of all fire extinguishing equipment corresponding to the surrounding fire extinguishing area, it is possible to achieve precise handling of fires at different stages of development and improve fire extinguishing efficiency.

[0084] In one exemplary embodiment, such as Figure 3 As shown, a method for obtaining fire control strategies for wine cellars is provided, which includes the following steps:

[0085] (1) Multimodal data acquisition: acquire multimodal data in the wine cellar environment, and acquire multiple grid areas based on the spatial structure information of the wine cellar, and acquire environmental parameters for each grid area; environmental parameters include baseline temperature and humidity and baseline dust level; construct a digital twin model of the wine cellar; map the multimodal data and environmental parameters to the digital twin model, and acquire the simulation results of the digital twin model; perform fusion filtering on the simulation results and temperature and humidity data in the multimodal data to obtain temperature and humidity compensation values; when it is determined that there is dust interference in the grid area based on visible light video data, perform attenuation processing on the multimodal data based on the baseline dust level to obtain dust compensation values; multimodal data includes visible light video data, infrared thermal imaging data, environmental sensor data, and combustible gas concentration data.

[0086] (2) Data feature extraction: Visual spatial features of target data are extracted through target detection network model, and temporal features of target data are extracted through machine learning network model.

[0087] (3) Feature fusion analysis: Visual spatial features and temporal features are fused to obtain fused features, and the fused features are fused and analyzed by a multimodal analysis model to obtain fire risk level and fire source location information.

[0088] (4) Fire source location and spread trend prediction: Obtain the motion vectors corresponding to the flame area and the temperature anomaly area, and obtain the flame trajectory based on visible light video data; perform principal component analysis on all motion vectors to obtain the flame spread direction, and obtain the flame spread speed based on the flame trajectory; discretize the digital twin model into multiple regular grids on a preset plane, and determine the state attributes corresponding to the regular grids; the state attributes include unburned, burning, burnt out, and non-combustible; obtain the ignition energy corresponding to the regular grids based on the flame spread direction and flame spread speed, and configure the grid ignition rules for the regular grids based on the ignition energy and state attributes to obtain the fire spread model; use the flame spread direction and flame spread speed as the model driving signal to perform multiple rounds of iterative solution on the fire spread model to obtain the fire spread areas corresponding to multiple time periods.

[0089] (5) Dynamic risk assessment: When the fire risk level is Level 1, an alarm message is triggered; when the fire risk level is Level 2, the target fire extinguishing equipment is determined according to the fire spread area, and the target fire extinguishing equipment is activated; when the fire risk level is Level 3, the fire extinguishing area is determined according to the fire spread area, and all fire extinguishing equipment corresponding to the fire extinguishing area is activated.

[0090] For example, taking a large baijiu (Chinese liquor) storage area of ​​5,000 square meters as an example, this storage area includes multiple functional areas such as a ceramic jar aging area, a bottled liquor storage area, a blending workshop, and passageways. To address the fire hazards unique to the baijiu storage environment, such as ethanol vapor accumulation, high temperature and humidity, and dust risks, a comprehensive multimodal sensor network was deployed. Twelve high-definition infrared PTZ cameras and eight infrared thermal imagers were deployed at key nodes in the storage area, forming complete visual monitoring coverage. Explosion-proof ethanol concentration sensors were installed in high-risk areas for alcohol evaporation, and temperature and humidity composite sensors and laser dust sensors were deployed in the overhead space. All sensor data is transmitted via the LoRa wireless protocol to four edge computing nodes deployed in different zones, responsible for real-time processing of the multimodal data for their respective areas.

[0091] During operation, the multimodal sensor network continuously collects environmental data. When an alcohol leak occurs in the blending workshop, the ethanol sensor detects that the concentration has risen to 25% LEL within 3 seconds, exceeding the preset threshold. Simultaneously, the thermal imager captures an anomaly where the local temperature suddenly rises from 25°C to 65°C. After the edge nodes perform timestamp alignment and Kalman filtering noise reduction on the multi-source data, the multimodal fusion analysis process is initiated. The visible light video stream is input into the improved YOLOv10s model, which accurately identifies a 2×3 meter open flame area with a confidence level of 0.89. The thermal imaging data is processed by a ResNet50 network to extract temperature distribution features, and the detected fire point temperature has reached 310°C. The time-series data of smoke concentration is analyzed by an LSTM network, showing that it conforms to the characteristic change pattern of a fire.

[0092] Based on the fusion analysis of multimodal data, an attention gating mechanism was used to weight and fuse various features, calculating a fire risk level of Level 1 with a risk score of 0.91 and a confidence level exceeding the alarm threshold of 0.7. A tiered response mechanism was then activated: first, the workshop's audible and visual alarms were triggered, and detailed alarm information including the fire location, temperature, and gas concentration data was pushed to the safety officer's mobile phone; simultaneously, the optimal fire extinguishing plan was calculated, dispatching the smart water cannon closest to the fire source (8 meters) for directional spraying, and, based on the fire spread direction predicted by optical flow, activating two water cannons in adjacent areas to form a flame-retardant defense line. Throughout the entire process, ventilation control, emergency lighting, and power management systems were simultaneously activated, achieving a comprehensive emergency response.

[0093] After six months of actual operation, the system demonstrated that, compared to traditional smoke alarm systems, fire detection accuracy increased from 78% to 96%, average response time decreased from 25 seconds to 8 seconds, and the average number of false alarms per month decreased from 5.2 to 0.3. Furthermore, the precise fire suppression resulted in a 77% reduction in the area affected by secondary water damage. Additionally, the system was specifically optimized for the high-temperature and high-humidity environment of the wine cellar. A humidity compensation factor was introduced into the LSTM model, and the sensors adopted an IP67 protection rating and passed a 96-hour salt spray test, ensuring long-term reliability in harsh environments. Through an edge-cloud collaborative mechanism, edge nodes upload feature data to the cloud platform every 24 hours for model optimization and weight updates, enabling the system to continuously learn and improve the accuracy and reliability of fire monitoring.

[0094] In this embodiment, multimodal data from the wine cellar environment is acquired, and drift compensation is performed on the multimodal data to obtain target data. The visual spatial features of the target data are extracted through a target detection network model, and the temporal features of the target data are extracted through a machine learning network model. The visual spatial features and temporal features are fused to obtain fused features, and the fused features are analyzed through a multimodal analysis model to obtain fire risk level and fire source location information. Based on the fire source location information and spatial structure information, a fire spread model is constructed, and the fire spread area is obtained based on the fire spread model. Based on the fire risk level and the fire spread area, a wine cellar fire control strategy is obtained. This can reduce the probability of false alarms, improve the accuracy of fire risk level detection, achieve accurate fire source location and accurate fire prediction, thereby improving the reliability of the fire control strategy.

[0095] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0096] Based on the same inventive concept, this application also provides a wine cellar fire control strategy acquisition device for implementing the wine cellar fire control strategy acquisition method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations of one or more wine cellar fire control strategy acquisition device embodiments provided below can be found in the limitations of the wine cellar fire control strategy acquisition method described above, and will not be repeated here.

[0097] In one exemplary embodiment, such as Figure 4 As shown, a wine cellar fire control strategy acquisition device is provided, comprising: a data compensation module 10, a feature extraction module 20, a feature fusion module 30, a fire prediction module 40, and a strategy acquisition module 50, wherein:

[0098] The data compensation module 10 is used to acquire multimodal data in the wine cellar environment and perform drift compensation on the multimodal data to obtain target data; the multimodal data includes visible light video data, infrared thermal imaging data, environmental sensor data, and combustible gas concentration data.

[0099] The feature extraction module 20 is used to extract the visual spatial features of the target data through the target detection network model and to extract the temporal features of the target data through the machine learning network model.

[0100] The feature fusion module 30 is used to fuse visual spatial features and temporal features to obtain fused features, and to perform fusion analysis on the fused features through a multimodal analysis model to obtain fire risk level and fire source location information.

[0101] The fire prediction module 40 is used to construct a fire spread model based on fire source location information and spatial structure information, and to obtain the fire spread area based on the fire spread model.

[0102] The strategy acquisition module 50 is used to acquire fire control strategies for the wine cellar based on the fire risk level and the area of ​​fire spread.

[0103] In an exemplary embodiment, the data compensation module 10 is further configured to acquire multiple grid regions based on the spatial structure information of the wine cellar, acquire environmental parameters for each grid region, including baseline temperature and humidity and baseline dust level; construct a digital twin model of the wine cellar; map the multimodal data and environmental parameters to the digital twin model, and acquire the simulation results of the digital twin model; and perform drift compensation on the multimodal data based on the simulation results.

[0104] In an exemplary embodiment, the data compensation module 10 is further configured to perform fusion filtering on the temperature and humidity data in the simulation results and multimodal data to obtain temperature and humidity compensation values; and when it is determined from the visible light video data that there is dust interference in the grid area, the multimodal data is attenuated based on the baseline dust level to obtain dust compensation values.

[0105] In an exemplary embodiment, the fire source location information includes a flame area and a temperature anomaly area; the fire prediction module 40 is further used to obtain the motion vectors corresponding to the flame area and the temperature anomaly area, and to obtain the flame trajectory based on visible light video data; to perform principal component analysis on all motion vectors to obtain the flame spread direction, and to obtain the flame spread speed based on the flame trajectory; to discretize the digital twin model into multiple regular grids on a preset plane, and to determine the state attributes corresponding to the regular grids; the state attributes include unburned, burning, burnt out, and non-combustible; to obtain the ignition energy corresponding to the regular grids based on the flame spread direction and the flame spread speed, and to configure grid ignition rules for the regular grids based on the ignition energy and the state attributes, thereby obtaining a fire spread model.

[0106] In an exemplary embodiment, the fire prediction module 40 is further configured to use the flame spread direction and flame spread speed as model driving signals to perform multiple rounds of iterative solutions to the fire spread model, thereby obtaining the fire spread areas corresponding to multiple time periods.

[0107] In an exemplary embodiment, the strategy acquisition module 50 is further configured to trigger an alarm message when the fire risk level is at the first level; determine the target fire extinguishing equipment based on the fire spread area when the fire risk level is at the second level, and control the target fire extinguishing equipment to start; and determine the surrounding fire extinguishing area based on the fire spread area when the fire risk level is at the third level, and control all fire extinguishing equipment corresponding to the surrounding fire extinguishing area to start.

[0108] The modules in the aforementioned wine cellar fire control strategy acquisition device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0109] In one exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 5As shown, the computer device includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. 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 in the non-volatile storage media. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, Near Field Communication (NFC), or other technologies. When the computer program is executed by the processor, it implements a method for acquiring a wine cellar fire control strategy. The display unit is used to form a visually visible image and can be a display screen, projection device, or virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0110] Those skilled in the art will understand that Figure 5 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0111] In one exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to perform the following steps: acquiring multimodal data in a wine cellar environment and performing drift compensation on the multimodal data to obtain target data; the multimodal data includes visible light video data, infrared thermal imaging data, environmental sensor data, and combustible gas concentration data; extracting visual spatial features of the target data through a target detection network model and extracting temporal features of the target data through a machine learning network model; fusing the visual spatial features and temporal features to obtain fused features, and performing fusion analysis on the fused features through a multimodal analysis model to obtain fire risk level and fire source location information; constructing a fire spread model based on the fire source location information and spatial structure information, and obtaining the fire spread area based on the fire spread model; and obtaining a wine cellar fire control strategy based on the fire risk level and the fire spread area.

[0112] In one embodiment, the process of a processor executing a computer program to perform drift compensation on multimodal data to obtain target data includes: acquiring multiple grid regions based on the spatial structure information of the wine cellar, and acquiring environmental parameters for each grid region; the environmental parameters include baseline temperature and humidity and baseline dust level; constructing a digital twin model of the wine cellar; mapping the multimodal data and environmental parameters to the digital twin model, and obtaining the simulation results of the digital twin model; and performing drift compensation on the multimodal data based on the simulation results.

[0113] In one embodiment, the drift compensation of multimodal data based on simulation results involved in the processor executing the computer program includes: performing fusion filtering on temperature and humidity data in the simulation results and multimodal data to obtain temperature and humidity compensation values; and, when it is determined from visible light video data that dust interference exists in the grid area, performing attenuation processing on the multimodal data based on the baseline dust level to obtain dust compensation values.

[0114] In one embodiment, the fire source location information includes a flame area and a temperature anomaly area. When the processor executes the computer program, it constructs a fire spread model based on the fire source location information and spatial structure information, including: acquiring motion vectors corresponding to the flame area and the temperature anomaly area, and acquiring the flame trajectory based on visible light video data; performing principal component analysis on all motion vectors to obtain the flame spread direction, and acquiring the flame spread speed based on the flame trajectory; discretizing the digital twin model into multiple regular grids on a preset plane, and determining the state attributes corresponding to the regular grids; the state attributes include unburned, burning, burnt out, and non-flammable; acquiring the ignition energy corresponding to the regular grids based on the flame spread direction and flame spread speed, and configuring grid ignition rules for the regular grids based on the ignition energy and state attributes, thus obtaining the fire spread model.

[0115] In one embodiment, the process of obtaining the fire spread area based on the fire spread model when the processor executes the computer program includes: using the direction and speed of flame spread as model driving signals, performing multiple rounds of iterative solution on the fire spread model to obtain the fire spread area corresponding to multiple time periods.

[0116] In one embodiment, the processor executing the computer program involves obtaining a fire control strategy for the wine cellar based on the fire risk level and the fire spread area, including: triggering an alarm message when the fire risk level is Level 1; determining the target fire extinguishing equipment based on the fire spread area and controlling the activation of the target fire extinguishing equipment when the fire risk level is Level 2; and determining the enclosed fire extinguishing area based on the fire spread area and controlling the activation of all fire extinguishing equipment corresponding to the enclosed fire extinguishing area when the fire risk level is Level 3.

[0117] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0118] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0119] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0120] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0121] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for obtaining fire control strategies for wine cellars, characterized in that, The method includes: Multimodal data under the wine cellar environment is acquired, and drift compensation is performed on the multimodal data to obtain target data; the multimodal data includes visible light video data, infrared thermal imaging data, environmental sensor data, and combustible gas concentration data; The visual spatial features of the target data are extracted using an object detection network model, and the temporal features of the target data are extracted using a machine learning network model. The visual spatial features and the temporal features are fused to obtain fused features, and the fused features are analyzed by a multimodal analysis model to obtain fire risk level and fire source location information; Based on the fire source location information and spatial structure information, a fire spread model is constructed, and the fire spread area is obtained based on the fire spread model. Based on the fire risk level and the fire spread area, obtain the fire control strategy for the wine cellar.

2. The method according to claim 1, characterized in that, The process of performing drift compensation on the multimodal data to obtain target data includes: Based on the spatial structure information of the wine cellar, multiple grid areas are obtained, and environmental parameters for each grid area are acquired; the environmental parameters include baseline temperature and humidity and baseline dust level. Construct a digital twin model of the wine cellar; The multimodal data and environmental parameters are mapped to the digital twin model, and the simulation results of the digital twin model are obtained. Drift compensation is performed on the multimodal data based on the simulation results.

3. The method according to claim 2, characterized in that, The drift compensation of the multimodal data based on the simulation results includes: The simulation results and the temperature and humidity data in the multimodal data are fused and filtered to obtain temperature and humidity compensation values; If dust interference is determined to exist in the grid area based on the visible light video data, the multimodal data is attenuated based on the baseline dust level to obtain a dust compensation value.

4. The method according to claim 2, characterized in that, The fire source location information includes the flame area and the temperature anomaly area; the step of constructing a fire spread model based on the fire source location information and the spatial structure information includes: Obtain the motion vectors corresponding to the flame region and the temperature anomaly region, and obtain the flame motion trajectory based on the visible light video data; Principal component analysis is performed on all motion vectors to obtain the flame spread direction, and the flame spread velocity is obtained based on the flame trajectory. The digital twin model is discretized into multiple regular grids on a preset plane, and the state attributes corresponding to the regular grids are determined; the state attributes include unburned, burning, burned out, and non-flammable; Based on the flame spread direction and the flame spread speed, the ignition energy corresponding to the regular grid is obtained, and the grid ignition rules are configured for the regular grid according to the ignition energy and the state attributes to obtain the fire spread model.

5. The method according to claim 4, characterized in that, The step of obtaining the fire spread area based on the fire spread model includes: Using the flame spread direction and flame spread speed as model driving signals, the fire spread model is solved in multiple rounds of iterations to obtain the fire spread areas corresponding to multiple time periods.

6. The method according to claim 1, characterized in that, The process of obtaining a fire control strategy for the wine cellar based on the fire risk level and the fire spread area includes: When the fire risk level is Level 1, an alarm message is triggered. When the fire risk level is Level 2, the target fire extinguishing equipment is determined based on the fire spread area, and the target fire extinguishing equipment is activated. When the fire risk level is Level 3, the fire suppression area is determined based on the fire spread area, and all fire suppression equipment corresponding to the fire suppression area is activated.

7. A device for acquiring fire control strategies for wine cellars, characterized in that, The device includes: The data compensation module is used to acquire multimodal data in the wine cellar environment and perform drift compensation on the multimodal data to obtain target data; the multimodal data includes visible light video data, infrared thermal imaging data, environmental sensor data, and combustible gas concentration data; The feature extraction module is used to extract the visual spatial features of the target data through the target detection network model, and to extract the temporal features of the target data through the machine learning network model. The feature fusion module is used to fuse the visual spatial features and the temporal features to obtain fused features, and to perform fusion analysis on the fused features through a multimodal analysis model to obtain fire risk level and fire source location information; The fire prediction module is used to construct a fire spread model based on the fire source location information and spatial structure information, and to obtain the fire spread area based on the fire spread model. The strategy acquisition module is used to acquire the fire control strategy for the wine cellar based on the fire risk level and the fire spread area.

8. 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 method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images