A dynamic feedback evaluation method for compressed air foam extinguishing

Abstract

The application discloses a compressed air foam fire extinguishing dynamic feedback evaluation method and belongs to the technical field of fire safety. Multi-source fire scene data is collected, time alignment and space registration are completed, foam coverage state and heat decay potential state are coupled, three stages of flame control, heat suppression and coverage preservation are divided, and a rekindling risk index is calculated. A pixel-level supplementary spraying demand diagram is constructed to realize fixed-point closed-loop supplementary spraying, and a parameter online calibration and abnormal degradation bottom mechanism are matched. The method can predict the rekindling risk in advance, improve the accuracy of supplementary spraying, reduce the consumption and disposal cost of fire extinguishing agents, and improve the efficiency and safety and reliability of fire extinguishing operations in petrochemical and storage scenes.

Description

Technical Field

[0001] This invention belongs to the field of fire safety technology, specifically relating to a dynamic feedback evaluation method for compressed air foam fire extinguishing, which can be applied to intelligent control of foam fire extinguishing, early warning of reignition risk, and optimization of precise supplementary spraying in scenarios such as petrochemical storage tanks and warehouse stacks. Background Technology

[0002] Compressed air foam fire extinguishing technology, with its multiple mechanisms of action including cooling, asphyxiation, heat insulation, and coverage, has been widely applied in scenarios such as surface fires in petrochemical storage tanks, warehouse stacks, underground confined spaces, and industrial plant surfaces. Compared to traditional water-based fire extinguishing systems, compressed air foam offers greater adjustability in terms of gas-liquid ratio, foam diameter distribution, foam layer thickness, and liquid phase precipitation rate. Therefore, its extinguishing effect depends not only on the total spray volume but also on the dynamic coupling relationship between the spraying, coverage, and decay phases. From a microscopic perspective, the heat flux in the fire drives the foam film to drain and rupture, leading to increased foam layer porosity and decreased local coverage continuity. Simultaneously, the release of residual heat and the escape of combustible vapors in the combustion zone create a time-varying feedback, meaning there is no one-to-one correspondence between surface coverage and the elimination of reignition risk. The observation methods commonly used in existing projects mainly rely on the disappearance of visible flames, the decrease in surface temperature, or the replenishment of spray at fixed intervals. However, these methods essentially compress the continuously evolving fire extinguishing process into a small number of discrete judgment nodes, making it difficult to characterize the relationship between the integrity of the foam layer, the rate of thermal decay, and the latent reignition trend. This can easily lead to delayed reignition prediction, waste of extinguishing agents, low on-site handling efficiency, and even cause secondary fires to escalate.

[0003] Current technologies cannot couple the identification of localized foam rupture and residual heat risk, resulting in delayed reignition prediction and a high risk of secondary fires escalating, causing significant economic losses and safety risks. Furthermore, the assessment and re-spray control are disconnected, leading to significant waste of agents through inefficient spraying and high disposal costs. The lack of phased judgment criteria and the reliance on single thresholds can easily result in misjudging fire extinguishing completion, leading to low operational efficiency and poor quality consistency. These problems directly restrict the effectiveness and safety of compressed air foam fire suppression systems and are urgent issues to be addressed in high-risk fire suppression scenarios such as petrochemical plants and warehouses. Summary of the Invention

[0004] The purpose of this invention is to provide a dynamic feedback assessment method for compressed air foam fire extinguishing, which solves the problems in the prior art that make it difficult to reconstruct the dynamic relationship between foam layer integrity and residual heat release from multi-source observation data, that the assessment results cannot be fed back to the supplementary spray control in a timely manner, and that there is a lack of staged judgment criteria for flame control stage, heat suppression stage and coverage stage, thereby realizing the early prediction of reignition risk and closed-loop control of compressed air foam spraying process.

[0005] To achieve the above objectives, the present invention provides a dynamic feedback evaluation method for compressed air foam fire extinguishing, deployed in a fire extinguishing system including an infrared thermal imager, a visible light camera, environmental sensors, a foam spraying actuator, and a computer device, wherein the computer device performs the following steps: Step S1, Multi-source Fire Data Acquisition and Unified Preprocessing: Infrared thermal images, visible light images, spray condition data, and environmental disturbance data are acquired during the compressed air foam extinguishing process. The acquired data undergoes unified time alignment and spatial registration to obtain multi-source fire data corresponding to the same control cycle. For condition data with sampling frequencies higher than the control cycle, interval statistics are performed; for image data with sampling frequencies lower than the control cycle, time-sequence-preserving periodic correspondence is performed. The visible light image is reprojected onto the infrared thermal image coordinate system using a pre-calibrated mapping relationship, ensuring that the foam surface image and the infrared thermal image correspond in the same spatial coordinates. Outlier filtering is performed for continuous numerical data in pressure, flow rate, and environmental disturbances; neighborhood replacement is performed for saturated pixels in the thermal image; for continuous image gaps, interpolation is performed to fill in gaps within the allowable range, and an anomaly flag is set and a degradation strategy is triggered when the gaps exceed the allowable range.

[0006] Step S2, Foam Cover Integrity Status Construction: Based on the multi-source fire data, a foam area mask is extracted. The foam coverage rate and its changes are calculated based on the foam area mask. The degree of foam layer damage is calculated based on the correspondence between the edge of the foam area mask and the high-temperature exposed area in the infrared thermogram, thereby constructing a coverage status characterizing the foam coverage integrity. This coverage status reflects the overall coverage level, coverage change trend, and local edge instability of the foam layer within the monitoring area. When the foam coverage rate is lower than a preset coverage lower limit, the current coverage status can be directly determined as coverage failure.

[0007] Step S3, Thermal Attenuation Potential State Construction: Based on the infrared thermal image, high-temperature hotspot regions are extracted, and thermal attenuation potential states are constructed according to the area distribution of the high-temperature hotspots, the spatial temperature gradient, and the changes in the average temperature of the high-temperature region between adjacent control cycles. The thermal attenuation potential state is used to characterize the intensity of residual heat release and whether the heat core has entered a stable attenuation process. It can further generate corresponding thermal attenuation trend quantities to reflect the upward or downward trend of thermal risk between adjacent control cycles.

[0008] Step S4, phased risk assessment: Based on the visible light image, flame visibility features are extracted to generate a flame visibility index. The fire extinguishing process is then divided into flame control, heat suppression, and coverage maintenance stages based on the flame visibility index and the heat decay potential. According to the current stage, the degree of foam layer damage, heat decay potential, foam coverage status, environmental disturbance data, and heat decay trend are weighted and fused to obtain a reignition risk index. This phased assessment highlights different control objectives for each fire extinguishing stage, such as flame suppression, heat core decay, and coverage maintenance.

[0009] Step S5, Refilling Decision Generation and Closed-Loop Control Mapping: Based on the reignition risk index and the refilling demand map constructed from local thermal risk characteristics, edge damage spatial characteristics, and foam decay spatial characteristics, multiple high-demand areas are sorted by risk intensity, and the refilling area and intensity are determined sequentially. The compressed air foam injection mechanism is then driven to perform fixed-point closed-loop refilling. The refilling demand map describes the distribution of local refilling demand in spatial units, and is used to further map the global risk assessment results into local fixed-point refilling actions. After the refilling action is executed, data is re-collected in the next control cycle, and the assessment and refilling decision are repeated to form a closed loop. When an observation anomaly or execution anomaly reaches a preset condition, the most recent effective risk index is inherited and the conservative refilling rule is switched to. When all sensing fails or the injection mechanism experiences a safety malfunction, a safety interlock is triggered, automatic refilling is stopped, and a manual takeover alarm is output.

[0010] Step S6, Model Training, Parameter Calibration and Online Update: Construct an offline sample set with tags indicating whether reignition will occur within a future preset observation period based on physical simulation samples and control experiment samples; Use a joint objective function to offline calibrate the stage threshold, stage risk weight, and fusion weight in the respray demand graph; When the accumulated online task data reaches a preset quantity, only the stage threshold, stage risk weight, and fusion weight in the respray demand graph are incrementally updated, while the equipment safety parameters remain frozen; When the reignition false alarm rate increases or the invalid respray rate does not meet the preset improvement conditions in the updated version, roll back to the previous stable version.

[0011] Step S7, Abnormal Degradation and Safety Backup: When an observational or execution anomaly reaches a preset condition, the most recent effective risk index is inherited and the conservative re-spraying rule is switched. Specifically, when thermal image frame loss, severe visible light obstruction, or abnormal spraying mechanism operation reaches a preset condition, the direct use of spatial assessment results is suspended, and fixed-cycle short-pulse re-spraying is driven by the most recent effective risk index; if the spraying mechanism malfunctions, automatic re-spraying is stopped and an alarm is output; if the observation returns to normal and the data quality conditions are met for a preset number of consecutive control cycles, the degradation mode is gradually exited and a smooth transition to the normal assessment process is achieved.

[0012] Furthermore, the injection condition data in step S1 may include at least the mixing pressure and foam liquid flow rate, and the environmental disturbance data may include at least the wind speed and wind direction; within each control cycle after unified time alignment, each modal data corresponds to the same evaluation and control time, thereby enabling the updates of coverage status, thermal status and reignition risk to be completed based on the same time scale.

[0013] Furthermore, in the foam region mask extraction in step S2, local texture uniformity is used to characterize the visual continuity of the foam surface to suppress misjudgment of foam caused by local smoke occlusion or surface reflection; the degree of foam layer damage mainly characterizes the proportion of high-temperature exposed pixels near the foam edge, which is used to identify local coverage instability in advance when the overall coverage is still high.

[0014] Furthermore, the thermal decay potential state in step S3 is not based solely on the average temperature, but also integrates the proportion of high-temperature regions, the spatial temperature gradient, and the upward trend of the average temperature in the high-temperature regions, thereby characterizing whether the thermonuclear is truly in a state of continuous decay; the thermal decay trend is used to describe the upward or downward trend of thermal risk between adjacent control cycles and serves as the input for subsequent reignition risk calculation.

[0015] Furthermore, the reignition risk assessment in step S4 can construct a state vector that includes at least foam coverage, foam layer breakage index, thermal decay potential index, normalized injection pressure, normalized foam liquid flow rate, normalized environmental wind disturbance intensity, coverage change, and thermal decay potential change; and set different risk factor weights for the flame control stage, heat suppression stage, and coverage preservation stage, so as to match the risk estimation target with the target of the current fire extinguishing stage.

[0016] Furthermore, in step S4, the reignition risk index can be compared with the stop spraying threshold, the warning threshold, and the forced spraying threshold to distinguish between the stop spraying, the fixed-point spraying preparation, and the high-priority spraying status; wherein, the stop spraying threshold is lower than the warning threshold, and the warning threshold is lower than the forced spraying threshold.

[0017] Furthermore, the replenishment demand map in step S5 describes the intensity of local replenishment demand in terms of spatial location. The local thermal risk feature is used to reflect the current local thermal load, the edge damage spatial feature is used to reflect the spatial relationship between local pixels and foam damage boundaries, and the foam attenuation spatial feature is used to reflect the attenuation area on the foam surface caused by drainage, evaporation, or thermal shock. After integrating the three, a spatial replenishment demand distribution that simultaneously reflects thermal risk, structural damage risk, and foam attenuation risk can be obtained.

[0018] Furthermore, in step S5, when the spraying mechanism is a fixed nozzle, the supplementary spray intensity command is preferably mapped to the valve opening duration; when the spraying mechanism is a steerable nozzle, the center of gravity of the supplementary spray area can be further mapped to the nozzle azimuth and pitch angles, which together with the supplementary spray intensity command constitute the spray control quantity.

[0019] Furthermore, the offline sample set in step S6 can be jointly constructed from physical simulation samples and control experiment samples, and the sample labels include at least whether reignition will occur within a future preset observation period; in the joint objective function, the reignition underreporting loss is used to constrain the situation where the reignition risk is not identified in advance, the over-spraying loss is used to constrain the resource waste caused by unnecessary spraying, and the identification lag loss is used to constrain the degree of lag between the warning time and the starting point of the thermal risk.

[0020] Furthermore, in step S7, when the observed anomaly or execution anomaly reaches the preset condition, the conservative supplementary spraying rule preferably generates a conservative supplementary spraying intensity based on the most recent effective risk index and executes a fixed-cycle short-pulse supplementary spraying; when the most recent effective risk index does not exist, a preset neutral risk value can be used as a substitute input to ensure that an executable control quantity can still be output in the degradation mode.

[0021] Furthermore, in an optional embodiment, the method further includes filtering outliers from continuous numerical data in pressure, flow, and environmental disturbances; performing neighborhood replacement on saturated pixels in thermal images; performing interpolation to complete images when continuous image loss does not exceed a preset number of frames; and setting an anomaly flag and triggering a degradation strategy when continuous loss exceeds a preset number of frames, thereby ensuring that state variables can still be continuously updated under short-term data anomaly conditions.

[0022] Furthermore, in an optional implementation, the online update objects in the method only include the stage threshold, stage risk weight, and supplementary spray fusion weight, while the safety parameters related to the equipment safety boundary remain frozen to avoid the online update affecting the safe operating range of the spraying equipment.

[0023] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement all the steps of the above-described dynamic feedback evaluation method for compressed air foam fire extinguishing.

[0024] Compared with existing technologies, this invention has the following advantages: By coupling the characterization of foam layer damage and thermal attenuation potential, this invention can identify local instability risks in advance when the overall coverage is high, significantly advancing the reignition warning window, reducing the probability of reignition accidents, and reducing the economic losses and safety risks caused by accidents; by achieving spatially precise targeted respraying through pixel-level respray demand maps, it significantly reduces the consumption of ineffective agents, lowers fire extinguishing costs, shortens the duration of high-risk situations, and improves fire extinguishing efficiency; by adopting a three-stage differentiated risk assessment to match the objectives of different fire extinguishing stages, it avoids misjudgment bias based on a single criterion, improves the consistency of fire extinguishing quality, and reduces the risk of reignition; with supporting abnormal degradation and safety fallback mechanisms, it can be implemented based on conventional sensing equipment and existing foam systems, adapting to existing system upgrades and new construction, with low engineering implementation costs. The entire solution can be widely adapted to the intelligent upgrade needs of compressed air foam fire extinguishing in various scenarios such as petrochemicals and warehousing. Attached Figure Description

[0025] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments are briefly described below. All the drawings are flowcharts related to the method flow of the present invention, used to illustrate the method steps of the present invention and their interrelationships.

[0026] Figure 1 The flowchart of the overall method in the embodiment of the present invention mainly shows the overall closed-loop process from multi-source data acquisition and preprocessing, state construction, phased risk assessment, spraying decision generation and control mapping, model calibration and online update, to abnormal degradation and safety fallback.

[0027] Figure 2 This is a flowchart of the data acquisition and state construction unit in an embodiment of the present invention. It mainly shows the process of multi-source data acquisition, time alignment, spatial registration and anomaly handling, as well as the construction of foam coverage state and thermal decay potential state.

[0028] Figure 3 This is a flowchart of the phased risk assessment and supplementary spraying decision unit in an embodiment of the present invention. It mainly shows the process of constructing state vectors, determining fire extinguishing stages, calculating phased risks, generating supplementary spraying demand diagrams, determining supplementary spraying areas, and outputting supplementary spraying intensity.

[0029] Figure 4 This is a flowchart of the model training, parameter calibration and version update unit in an embodiment of the present invention, mainly showing the process of offline sample construction, threshold and weight calibration, validation set evaluation, online incremental update and version rollback.

[0030] Figure 5 The flowchart of the abnormal degradation and safety fallback unit in this embodiment of the invention mainly shows the process of multi-source observation anomaly detection, actuator status judgment, conservative replenishment rule triggering, fault alarm output, and degradation mode exit. Detailed Implementation

[0031] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0032] Implementation Method 1: Overall Process

[0033] like Figures 1 to 5 As shown, the dynamic feedback assessment method for compressed air foam fire extinguishing provided in this embodiment includes seven collaborative stages: multi-source data acquisition and preprocessing, foam coverage integrity state construction, thermal attenuation potential state construction, phased risk assessment, supplementary spraying decision generation and control mapping, model calibration and online updating, and abnormal degradation and safety fallback. This method is executed by the processor of a computer device, which is communicatively connected to an infrared thermal imager, a visible light camera, a pressure sensor, a flow meter, a wind speed and direction sensor, and a compressed air foam spraying mechanism.

[0034] Step S1: Data Acquisition and Unified Preprocessing of Multi-Source Fire Sites

[0035] S101. Acquire multi-source data. The processor reads the temperature matrix output from the infrared thermal imager, the foam surface image output from the visible light camera, the foam liquid flow rate output from the flow meter, the mixed gas-liquid pressure output from the pressure sensor, and the environmental disturbance data output from the wind speed and direction sensor within each control cycle. The infrared thermal image frame is denoted as: , in, Indicates the first Temperature matrix for each sampling period; Indicates the number of pixels at the height of the thermal image; This indicates the number of pixels representing the width of the thermal image. In this embodiment, , .

[0036] Visible light images are denoted as: , in, Indicates the first RGB image with sampling periods; and These represent the image height and width, respectively. In this embodiment, , The three channels correspond to the red, green, and blue color components, respectively.

[0037] S102, Perform uniform time alignment.

[0038] Because infrared thermal images, visible light images, and pressure and flow data have different sampling frequencies, the processor uses the control cycle as a unified time reference. For pressure and flow data with sampling frequencies higher than the control cycle, the processor performs interval averaging within the current control cycle; for image data with sampling frequencies lower than or equal to the control cycle, the processor performs time-sequence-preserving cycle correspondence to form unified time-stamped data under the same control cycle.

[0039] For pressure and flow rate, the processor also normalizes them based on the equipment's safe operating range. In this embodiment, the mixed pressure operating range can be 0.65 MPa to 0.95 MPa, which is obtained through offline calibration during stable operation testing of the spraying equipment; the foam liquid flow rate operating range can be 18 L / min to 30 L / min, which is determined by the rated operating conditions of the spraying system and the actual spraying boundary. If an abnormal situation occurs during equipment calibration where the maximum and minimum values ​​are the same, the processor determines that the equipment calibration is abnormal and will not proceed to the subsequent evaluation process.

[0040] S103, Execution space registration and exception handling.

[0041] The processor uses a pre-calibrated mapping relationship to reproject the visible light image onto the infrared thermal imaging coordinate system, ensuring that the foam surface image and the thermal temperature image correspond pixel-by-pixel in the same spatial coordinate system. This mapping relationship can be obtained offline using a field calibration board and is permanently saved before deployment.

[0042] After spatial registration is completed, the processor performs sliding median filtering on continuous numerical data in pressure, flow, and environmental disturbances to remove impulse noise; for saturated pixels in thermal images, the average of the effective pixels in their neighborhood is used for replacement; for continuously missing frames, linear interpolation is performed to fill in the missing frames when the number of consecutive missing frames does not exceed a preset number. In this embodiment, the preset number of frames can be 2 frames, which is determined based on the buffer capacity of the thermal imaging device and the link stability test; when the number of consecutive missing frames exceeds the preset number, the processor sets an anomaly flag and triggers subsequent degradation strategies.

[0043] After processing in step S1, the data of different modes are mapped to a unified time scale and a unified spatial coordinate system. Subsequent state construction can synchronously use thermal imaging information, visible light information, jetting conditions and environmental disturbance information under the same control cycle.

[0044] Step S2: Constructing the Integrity State of Foam Coverage

[0045] S201. Extract the foam region mask. The processor converts the spatially registered visible light image to the HSV color space, extracts the luminance and saturation components, and combines local texture uniformity to determine the foam pixels. Local texture uniformity reflects the degree of grayscale dispersion of the foam surface within a local window, and its calculation relationship is as follows: , in, Indicates the first Pixels in each period Local texture uniformity at the location; Represented in pixels The standard deviation of gray levels within a local window centered on the target; This represents the upper bound of the texture dispersion calibration, which can be set to 40 in this embodiment. This value is obtained by offline calibration through statistical analysis of the foam coverage annotation samples. This indicates a limiting operation, used to restrict the result to between 0 and 1. The closer to 1, the more uniform the local texture, which better matches the visual characteristics of the foam-covered area.

[0046] Based on brightness, saturation, and texture uniformity, the processor generates a foam region mask. In this embodiment, the foam brightness threshold, foam saturation threshold, and foam texture uniformity threshold can be set to 0.62, 0.28, and 0.55, respectively. These thresholds are obtained offline by calibrating the foam coverage labeled samples in petrochemical shallow pool fire and warehouse stack surface fire control experiments, with the goal of maximizing the intersection-union ratio (IU). In general scenarios, adaptation calibration can be completed by using an IU grid search method based on the labeled samples.

[0047] To suppress isolated misjudgments caused by localized smoke obstruction, the processor can also perform small connected component removal and closing operations on the mask. In this embodiment, the small connected component area threshold can be set to 9 pixels, which is determined based on the statistical analysis of the minimum stable foam block area; the closing operation radius can be set to 2 pixels, which is determined based on the thermal image and visible light registration error and the requirement for smoothing foam edges.

[0048] S202. Calculate the foam coverage rate and the change in coverage rate.

[0049] The processor calculates the foam coverage based on the pixel percentage of the foam region mask within the monitored area: , in, Indicates the first The foam coverage rate for each control cycle ranges from 0 to 1. and These represent the height and width pixel counts of the monitored area in the thermal imaging coordinate system, respectively. Indicates the first Each period of pixels Whether a pixel is identified as a bubble pixel is determined by a value of 1, which indicates that it belongs to a bubble area, and a value of 0 indicates that it does not belong to a bubble area. This value directly reflects the proportion of the monitored area that is covered by bubbles.

[0050] The coverage difference between adjacent control cycles yields the coverage change, which is then normalized according to a preset maximum effective change amplitude. This maximum effective change amplitude can be determined offline based on the high quantile of the training samples. In this embodiment, this amplitude can be 0.20, derived from the 95th percentile of the single-cycle coverage change statistics. When there is no data from the previous cycle in the first cycle, the normalized result of the coverage change can be set to 0. When the effective pixel percentage of the image in the current cycle is less than 70%, the processor does not update the coverage change, but instead uses the result from the previous effective cycle and sets the data quality flag.

[0051] S203. Calculate the foam layer damage index.

[0052] The processor performs morphological gradient operations on the mask boundary of the foam region to obtain an edge set. It then calculates the proportion of edge pixels in the edge set that correspond to thermal images exceeding a preset high-temperature exposure threshold, reflecting the degree of high-temperature exposure near the foam edge. The calculation relationship is as follows: , in, Indicates the first The foam layer damage index for each control cycle typically ranges from 0 to 1. A higher value indicates a higher proportion of high-temperature pixels exposed near the foam edge. Indicates the first A set of edge pixels of a periodic foam region mask; Temperature matrix In pixels The temperature value at that location, in degrees Celsius; This represents the high-temperature exposure threshold, which can be taken as 180 degrees Celsius in this embodiment. This threshold is obtained by offline calibration based on the thermal image statistics under continuous heat release conditions. This indicates an indicator function that takes the value 1 if the condition within the parentheses is true, and 0 otherwise. Indicates the number of pixels at the edge; This represents a smoothing term to prevent the denominator from being 0. In this embodiment, it can be set to 1 to ensure that the value is stable when there are very few edge pixels.

[0053] When the foam coverage rate is lower than the preset coverage lower limit, the processor can directly determine the current coverage state as coverage failure. In this embodiment, the coverage lower limit can be 0.05, which is determined based on the test results of edge indicators losing stable physical meaning under extremely low coverage conditions.

[0054] Through step S2, the system not only obtains the overall coverage rate, but also the degree of edge damage, thus enabling it to distinguish between states where the overall coverage appears to still exist, but hot spots have already penetrated the edges.

[0055] Step S3: Construction of thermal decay potential state

[0056] S301, Extract the high-temperature hot spot area.

[0057] The processor performs threshold segmentation on the temperature matrix to obtain a high-temperature region mask. The hotspot threshold is used to distinguish between regions with normal temperature rise and hot core regions with a risk of continuous heat release. In this embodiment, the hotspot threshold can be set to 200 degrees Celsius, which is obtained through offline calibration of the thermal image distribution during the high-risk phase of the control experiment.

[0058] Subsequently, the processor calculates the average temperature of the high-temperature region. When there is no high-temperature region in the current cycle, the processor marks the current cycle as a high-temperature region-free cycle and uses the most recent valid high-temperature cycle to replace it in subsequent temperature rise reversal calculations. If there is no high-temperature region for several consecutive cycles, the temperature rise reversal factor is set to a low value. S302, calculate the hotspot area ratio and temperature gradient energy.

[0059] The processor calculates the hot spot area ratio based on the percentage of pixels above the hot spot threshold: , in, Indicates the first The ratio of high-temperature hot spot area in each control cycle reflects the proportion of the area in the current monitoring region that is in the state of high-temperature hot spot. The hotspot threshold is represented, which can be 200 degrees Celsius in this embodiment; the meanings of the other symbols are the same as above. The hotspot area ratio can be further normalized to the range of 0 to 1 according to preset upper and lower bounds. The upper bound can be determined jointly based on the 95th percentile value of the high-risk stage in the training samples and the engineering upper limit.

[0060] The processor also calculates the temperature gradient energy based on the intensity of the temperature difference between adjacent pixels in the thermal image in the horizontal and vertical directions. The temperature gradient energy reflects the sharpness of the thermal boundary; the sharper the edge of a local hot spot, the more concentrated the local heat core, and the greater the likelihood of it penetrating the foam layer. Differences at image edges can be filled using mirroring; when the temperature of an adjacent pixel is missing, the nearest valid pixel is used as a replacement.

[0061] S303. Calculate the thermal decay potential index and its change.

[0062] To simultaneously characterize the current residual heat release intensity and the trend of heat core resurgence, the processor integrates the hot spot area ratio, temperature gradient energy, and temperature rise reversal factor into a thermal decay potential index: , in, Indicates the first The thermal decay potential index for each control cycle ranges from 0 to 1. The larger the value, the stronger the residual heat release and the less the heat core has entered a stable decay state. Indicates the ratio of hot spot area The result after normalization; This represents the result after normalizing the temperature gradient energy. Indicates the temperature rise reversal factor; , and These represent the fusion weights of the three parts, respectively, ensuring that the sum of the three is 1. In this embodiment, the three weights can be 0.35, 0.30, and 0.35, respectively. This combination is determined by searching the data grid within the observation window before reignition occurs in the simulation sample and the control experiment sample.

[0063] The temperature rise reversal factor is used to describe whether the average temperature in the high-temperature region has rebounded. In this embodiment, the noise reduction shift term can be 3 degrees Celsius, and the upper bound of the temperature rise reversal normalization can be 40 degrees Celsius. When there is no high-temperature region in the current cycle or no high-temperature region in the previous effective cycle, the processor uses the most recent effective high-temperature cycle instead. If there is no high-temperature region for three consecutive cycles, the temperature rise reversal factor is set to 0.

[0064] The processor also calculates the change in thermal decay potential based on the difference between the thermal decay potential indices of adjacent control cycles. To suppress short-time noise, the processor can perform short-window smoothing on the thermal decay potential indices before calculating the difference.

[0065] Through step S3, the system no longer relies solely on the average temperature to determine whether fire extinguishing is complete, but simultaneously considers the area of ​​high-temperature hot spots, spatial thermal gradient, and temperature rise reversal trend to provide a more detailed characterization of the residual heat release state.

[0066] Step S4: Phased Risk Assessment

[0067] S401. Construct the state vector.

[0068] The processor constructs a state vector for the current cycle risk assessment: , in, Indicates the first The state vector of each control cycle; Indicates foam coverage; Indicates the foam layer damage index; Indicates the thermal decay potential index; This represents the normalized mixed pressure; This represents the normalized foam liquid flow rate; This represents the normalized intensity of environmental wind disturbance. This represents the normalized value of the change in coverage. Represents the normalized value of the change in thermal decay potential; symbol This indicates transpose.

[0069] The intensity of environmental wind disturbance is characterized by both wind speed and wind direction, and the calculation relationship is as follows: , in, Indicates the first Normalized environmental wind disturbance intensity for each control cycle; This indicates the average wind speed for the current period, expressed in meters per second. This indicates the maximum allowable calibrated wind speed in the target scenario, which can be taken as 4 meters per second in this embodiment. This value is determined by the on-site wind disturbance limit test. This represents the angle between the current wind direction and the direction of the main jet axis, with a value ranging from 0 to... ; The lower limit of the directional weight is represented, which can be taken as 0.35 in this embodiment. This parameter is obtained through offline calibration based on the influence of crosswinds and headwinds on the stability of the foam edge. The formula reflects that the greater the wind speed and the closer the wind direction is to vertical or opposite, the higher the wind disturbance intensity.

[0070] S402, Perform phased identification.

[0071] The processor extracts the flame color response and temporal flicker response from the visible light image and fuses them into a flame visibility index: , in, Indicates the first The flame visibility index for each control cycle, ranging from 0 to 1; This represents the normalized value of the flame color response; This represents the normalized value of the flame flicker response; and This represents the weighting coefficients, where the sum of the two is 1. In this embodiment, the two can be 0.6 and 0.4 respectively, and these values ​​are obtained through offline calibration based on the recognition effect of open flame visibility and flicker stability.

[0072] The flame color response is constructed based on the proportion of warm-colored pixel area, while the flame flicker response is constructed based on the fluctuation of the warm-colored region area sequence over multiple consecutive control cycles. The flicker analysis window length in this embodiment can be 5 frames, a value determined statistically based on the short-term flame fluctuation frequency; the smoothing term in the flicker denominator can be... This is used to avoid dividing by zero when the average value is close to 0.

[0073] The stage determination rules are as follows: when the flame visibility index is not lower than the flame visibility threshold, it is determined to be the flame control stage; when the flame visibility index is lower than the flame visibility threshold and the heat decay potential index is not lower than the heat decay threshold, it is determined to be the heat suppression stage; when the flame visibility index is lower than the flame visibility threshold and the heat decay potential index is lower than the heat decay threshold, it is determined to be the protection stage. In this embodiment, the flame visibility threshold can be taken as 0.12, and the heat decay threshold can be taken as 0.35. These two thresholds are obtained by joint calibration based on manual annotation and reignition prediction results.

[0074] S403, Calculate the reignition risk index.

[0075] After identifying the current stage, the processor uses the corresponding risk weight to perform weighted fusion of the foam layer damage index, thermal decay potential index, insufficient foam coverage, environmental wind disturbance intensity, and the upward trend of thermal decay potential to obtain the reignition risk index: , in, Indicates the first The resurgence risk index for each control cycle ranges from 0 to 1. Indicates the current stage category. Corresponding to the flame control stage, Corresponding to the heat suppression stage, Corresponding to the protection and maintenance stage; to These represent the weights of the foam layer damage index, thermal decay potential index, insufficient foam coverage, environmental wind disturbance intensity, and the upward trend of thermal decay potential in the current stage, respectively, satisfying that the sum of the five weights in the same stage is 1. This indicates that only an upward trend in thermal risk will be positively penalized.

[0076] In this embodiment, the weights for the flame control stage can be 0.16, 0.22, 0.34, 0.12, and 0.16; the weights for the heat suppression stage can be 0.26, 0.31, 0.20, 0.09, and 0.14; and the weights for the protection stage can be 0.30, 0.28, 0.22, 0.08, and 0.12. These parameters are obtained by calibrating the minimum classification error of tags indicating whether reignition will occur within the next 30 seconds in simulation and experimental data.

[0077] The processor then compares the reignition risk index with the stop-spray threshold, the warning threshold, and the forced spray threshold to determine whether to stop spraying, enter the designated spraying preparation phase, or enter the high-priority spraying state. In this embodiment, the three thresholds can be set to 0.15, 0.22, and 0.45, respectively. These values ​​are obtained by comprehensively compromising the reignition false negative rate, the invalid spraying rate, and the identification lag. Specifically, the stop-spray threshold is selected with the goal of having a low invalid spraying rate; the warning threshold is selected with the constraint that the false negative rate is no higher than 10%; and the forced spraying threshold is selected with the goal of having a high recall rate for high-risk samples.

[0078] If any input component is missing and cannot be recovered in the current cycle, the system inherits the stable value from the previous cycle and adds a missing penalty. In this implementation, the missing penalty can be 0.05, which is set according to the conservative safety requirements under abnormal data; if the missing component is missing for more than three consecutive cycles, conservative supplementary spraying is directly triggered.

[0079] Step S5: Decision generation and closed-loop control mapping for supplementary spraying

[0080] S501, Generate the respray requirement diagram.

[0081] Based on the temperature distribution, spatial location of foam damage edges, and foam surface brightness attenuation information in the infrared thermal image, the processor constructs a pixel-level respray requirement map: , in, Indicates the first Each control cycle pixel The intensity of the need for respraying at the location; This represents the normalized temperature value; Indicates the normalized proximity of the damaged edge; This represents the local evaporation characterization value; , and The fusion weights of the three factors are such that their sum equals 1. In this embodiment, they can be 0.45, 0.35, and 0.20 respectively. This combination is obtained through offline calibration by compromising between the accuracy of the supplementary spray positioning and the total amount of supplementary spray.

[0082] The normalized temperature value is obtained by normalizing the thermal image temperature according to preset upper and lower boundaries. In this embodiment, the lower boundary of temperature normalization can be 60 degrees Celsius, and the upper boundary can be 350 degrees Celsius. This range is determined based on the statistical analysis of the sensitive interval of the medium and high temperature region for the supplementary spraying decision.

[0083] The proximity of the damaged edge is constructed using the shortest distance from the pixel to the edge of the bubble, and its core relationship is: , in, Represents pixels Normalized proximity of the damaged edge at the location; Represents pixels The shortest Euclidean distance to the set of bubble edges, in pixels; The proximity attenuation scale can be 4 pixels in this embodiment. This parameter is determined based on the spray radius and the edge thermal diffusion range. This represents a candidate thermal risk region consisting of a high-temperature area and a certain number of pixels extending outward. In this embodiment, the outward radius can be 3 pixels, and this value is determined by statistics of the hot spot expansion boundary. As an indicator function, edge proximity is retained only within candidate thermal risk regions. This relationship suggests that the closer a pixel is to the damaged edge and within the thermal risk neighborhood, the greater its need for re-spraying.

[0084] The local evaporation characterization value is approximated by the brightness decay of the foam surface over two consecutive frames. Essentially, it reflects the rapidly darkening areas on the foam surface caused by drainage, evaporation, or localized thermal shock. In this embodiment, the upper bound of the brightness decay normalization can be taken as 0.35, which is obtained through offline calibration based on statistical data of foam surface brightness change samples.

[0085] When a pixel lacks both visible light and thermal image effective values, the processor sets the pixel's spraying requirement to 0 to avoid accidental spraying in invalid areas. If the effective pixel ratio of the entire image is less than 70%, the spatial spraying requirement map will not be used, and global conservative spraying control will be switched to instead.

[0086] S502, Extract the respray area and generate the respray intensity command.

[0087] The processor performs threshold segmentation on the re-spray demand map to extract high-demand connected regions. In this embodiment, the high-demand threshold can be set to 0.55, which is determined offline by balancing re-spray positioning error and missed selection rate. Subsequently, the processor counts the number of high-demand connected regions with an area not less than the minimum effective area, sorts them by area from largest to smallest, and selects the top few regions as the re-spray region set. In this embodiment, the maximum number of regions participating in the decision-making process can be set to 3, with this upper limit determined based on the single-cycle response capability of the actuator; the minimum effective area can be set to 1% of the total number of pixels, and this threshold is determined offline based on the minimum executable re-spray patch area.

[0088] The processor further calculates the total area ratio of high-demand regions and generates supplementary spray intensity instructions based on the global reignition risk index: , in, Indicates the first The intensity command for each control cycle of the additional spraying; This represents the minimum supplementary spray intensity, which can be taken as 0.2 in this embodiment. This value is determined based on the spraying boundary that ensures the minimum effective supplementary spraying effect. This represents the maximum respray intensity, which can be taken as 1.0 in this embodiment. This value is determined based on the maximum safe spraying time before significant foam erosion damage occurs. and The weighted average of the resurgence risk index and the total area ratio of high-demand regions is 0.6, which can be 0.4 and 0.4 respectively in this implementation. This indicates the proportion of the total area in high-demand regions; This indicates a limit operation.

[0089] when Below the stop spraying threshold and At lower levels, the processor directly sets the respray intensity to 0, indicating that no respray is needed. In this embodiment, A lower criterion of 0.05 can be used, a value determined based on empirical testing that suggests localized, scattered, high-demand patches are not worth driving injections individually. When When the pressure is not lower than the forced supplementary spray threshold, at least one supplementary spray with an intensity not lower than the minimum supplementary spray intensity shall be performed. When the current pressure of the spraying mechanism is lower than the safety lower limit, the processor shall not increase the supplementary spray intensity, but instead output a device alarm and perform minimum safe spraying. In this embodiment, the safety lower limit pressure can be 0.60 MPa, which is obtained by calibrating the minimum stable bubble output pressure of the spraying mechanism.

[0090] S503, Execute supplementary spray control mapping.

[0091] When the spraying mechanism is a fixed nozzle, the processor maps the supplementary spray intensity command to the valve opening duration: , in, Indicates the first Valve opening duration per control cycle, in seconds; This represents the minimum effective pulse width, which can be 0.8 seconds in this embodiment. This value is obtained by offline calibration of the valve opening and closing test. This indicates the maximum opening time in a single cycle, which can be taken as 4.0 seconds in this embodiment. This value is determined by the maximum allowable supplementary spray amount in a single cycle and the valve's continuous opening capability. This is a command to increase the spray intensity. The average valve action delay in this embodiment can be 0.12 seconds. This delay is measured by bench opening and closing tests, and the processor compensates for this delay relative to the evaluation result when issuing the control.

[0092] When the spray mechanism uses a steerable nozzle, the processor maps the center of gravity of the supplementary spray area to the nozzle target direction. A three-dimensional coordinate system is established with the nozzle mounting point as the origin, where... The axis is aligned with the initial orientation of the nozzle. The axis is horizontal. The axis is vertical. After calibration from the thermal image field of view to the actual plane, the spatial coordinates corresponding to the centroid of the spraying area are: Then the azimuth and elevation angles can be expressed as: , , in, Indicates the number of nozzles The target azimuth angle for each control cycle; Indicates the number of nozzles The target pitch angle for each control cycle; , and This indicates the spatial coordinates of the centroid of the supplementary spraying area in the nozzle coordinate system, and the unit can be meters; This represents the arctangent operation. The zero angle and proportional coefficient are obtained from the nozzle installation calibration. In this embodiment, the maximum angular velocity of the nozzle can be 60 degrees per second, the minimum resolvable angle can be 1 degree, and when the control cycle is 1 second, a single azimuth adjustment is allowed to be no more than 45 degrees. This range is determined based on the limits of the actuator motor and the smoothness of control. If the target angle exceeds the range achievable in a single cycle, it is approximated along the shortest path in multiple control cycles. When an observed anomaly or execution anomaly reaches a preset condition, the most recent effective risk index is inherited and the conservative supplementary spraying rule is switched. When all sensors fail or a safety fault occurs in the spraying mechanism, a safety interlock is triggered, automatic supplementary spraying is stopped, and a manual takeover alarm is output.

[0093] In the next control cycle after the respraying action is executed, the processor re-acquires data and repeats steps S1 to S5, thereby forming a closed loop of acquisition, evaluation, respraying, and re-evaluation.

[0094] Step S6: Model training, parameter calibration, and online updates

[0095] S601. Construct an offline sample set.

[0096] The processor constructs an offline sample set tagged with reignition based on physical simulation samples and control experiment samples. Simulation samples are obtained by changing the initial fire surface temperature, foam layer thickness, wind speed, injection pressure, and fuel type; control experiment samples are obtained through real or semi-real fire extinguishing experiments. Each sample sequence records at least whether reignition occurs within a preset observation period, the reignition time, and whether stability is restored after re-spraying. The reignition determination criterion is that the visible light flame visibility index exceeds the open flame threshold and the heat decay potential index continues to rise.

[0097] In this embodiment, a total of 80 independent task sequences were collected in the control experiment, with each condition repeated 4-6 times. These included 48 shallow pool fire sequences and 32 stack surface fire sequences. The statistical caliber used was the task-level mean. Operating variables included injection pressure (0.65 MPa to 0.90 MPa), flow rate (18 L / min to 30 L / min), crosswind (0 m / s to 3.6 m / s), and main injection duration (25 s to 45 s). 22 task sequences exhibited localized thermonuclear resurgence after open flame suppression, serving as samples for reignition risk labeling.

[0098] S602, Perform offline calibration.

[0099] The processor employs a joint objective function to search and validate the stage threshold, stage risk weight, and supplementary spray fusion weight: , in, Indicates the total loss; This represents the percentage of reignition losses that were missed during the pre-set observation period, but for which no risk warning has been triggered at present. This indicates the loss from excessive re-spraying, which is the proportion of medium- to high-intensity re-spraying triggered despite no risk of reignition. This indicates the identification lag loss, which is the normalized time difference between the early warning time and the starting point of the thermal risk. , and The values ​​represent the weights of the three types of losses. In this embodiment, the weights can be 0.5, 0.2, and 0.3, respectively. These values ​​reflect the design goal of prioritizing the suppression of false negatives while also taking into account overshooting and identification lag.

[0100] When there are no reignition samples in the validation set, the reignition false negative loss can be replaced by the statistical value of the previous batch containing reignition samples; when the number of samples of a certain type is too small, category reweighting can be used. The stage threshold and action threshold obtained after calibration can be used as parameters for the initial deployment version.

[0101] S603, Perform online updates and version rollbacks.

[0102] After online deployment, the processor adds the task data to the update pool after each complete firefighting mission. When the cumulative number reaches a preset threshold, the new parameters are tested using a reserved validation set. In this embodiment, the update threshold can be set to 50 task sequences, a value determined by a trade-off between parameter stability and update frequency.

[0103] Online updates are limited to stage thresholds, stage risk weights, respray demand map fusion weights, and reignition risk-to-area ratio fusion weights. Pressure range, flow range, safety lower limit, minimum effective pulse width, maximum start duration, and execution delay parameters related to equipment safety boundaries remain frozen and are not included in online updates.

[0104] If the updated version meets the requirements of no increase in the re-ignition false alarm rate and a decrease in the invalid re-spray rate of no less than 2%, then the main version is updated; otherwise, it is rolled back to the previous stable version. The 2% improvement condition is set based on the requirement to avoid frequent small fluctuations in updates during actual operation and maintenance. If a task sequence meets the requirement that continuous sensor anomalies exceed 10% of the control cycle or that an actuator failure causes the action to not actually take effect, then the sequence will not enter the parameter update pool, but only the fault analysis pool.

[0105] Step S7: Abnormal Degradation and Safety Net

[0106] S701, Detect abnormal status.

[0107] The processor continuously monitors thermal imaging, visible light, and nozzle operating conditions. It enters degradation mode when thermal image distortion occurs for three consecutive cycles, visible light is severely obstructed, or nozzle pressure is abnormal. Thermal image distortion can include a saturated pixel ratio exceeding 40% or more consecutive frame drops exceeding two frames; severe visible light obstruction can include an effective pixel ratio below 50% for at least two consecutive control cycles; and abnormal pressure can include a current pressure below 0.60 MPa or pressure fluctuations exceeding 0.12 MPa within three control cycles. These thresholds were all obtained through offline calibration via equipment testing and abnormal scenario experiments.

[0108] S702, Implement conservative touch-up spraying rules.

[0109] In degradation mode, the processor no longer relies on the space replenishment demand map, but instead inherits the most recently effective risk index to generate a conservative replenishment intensity: , in, Indicates the first Conservative spraying intensity under the degradation mode of each control cycle; This represents the reignition risk index for the most recent effective cycle; the constant 0.4 is the basic respray intensity and lower limit under degradation mode, which is determined offline based on tests to maintain basic coverage capability without spatial positioning; the upper limit 0.8 is determined based on safety requirements to avoid excessive erosion of the foam layer under degradation mode.

[0110] For fixed nozzles, the on-time in degradation mode is: , in, Indicates the duration the fixed nozzle is on in degradation mode; and The meaning is the same as before. If the most recent effective risk index does not exist, then the default is... This value is used as a neutral risk value; if the injection mechanism malfunctions, automatic re-spraying will stop, and only the alarm output will be retained; if the observation returns to normal and the data quality conditions are met for two consecutive cycles, the degradation mode will be exited and the process will return to step S1.

[0111] After step S7, the system has a switchable safety fallback mechanism between normal mode and degradation mode, so as to maintain control continuity even under conditions of multi-source observation anomalies or execution fluctuations.

[0112] Implementation Method 2: Dynamic Feedback Assessment and Supplementary Spraying Control in Petrochemical Shallow Pool Fire Scenarios

[0113] This embodiment describes a shallow pan fire scenario in a petrochemical plant area. In this scenario, conventional treatment typically involves re-spraying at fixed intervals after the main spray, with the judgment mainly based on whether the visible flame has disappeared and manual thermal observation. When the open flame is briefly suppressed, the edges of the foam layer are prone to local damage due to radiant heat and lateral wind disturbance, which can then form reignition points in high-temperature areas. Manual observation often only allows for re-spraying after the open flame reappears, resulting in a delayed response.

[0114] In this embodiment, the fire plate diameter is 2.4 m; the thermal image resolution is 64×64; the visible light resolution is 640×640; the thermal image sampling period is 0.5 s; the visible light sampling period is 0.5 s; the pressure and flow rate sampling period is 0.1 s; and the unified control period is 1 s. The foam spraying pressure operating range is 0.65 MPa to 0.95 MPa, with the main spraying pressure set at 0.80 MPa in this scenario; the foam liquid flow rate operating range is 18 L / min to 30 L / min, with the main spraying pressure set at 24 L / min; the supplementary spraying intensity range is 0 to 1, corresponding to a valve opening time of 0.8 s to 4 s; the average valve action delay is 0.12 s; the lower safety limit pressure is 0.60 MPa; the normal wind speed is 1.2 m / s, and the peak value under impact conditions is 3.6 m / s. These parameters were obtained through shallow pan fire plate testing, valve opening and closing testing, and offline calibration of the on-site operating conditions.

[0115] In the 36th control cycle, the main injection has ended, and the system has entered the maintenance and observation period. In step S1, the processor receives the current cycle data: the average temperature of the high-temperature area is 214 degrees Celsius, and the average temperature of the high-temperature area in the previous cycle was 207 degrees Celsius; the number of high-temperature hotspot pixels is 328; the number of foam coverage pixels is 3215; the number of foam edge pixels is 412; the number of pixels in the edge that are higher than the high-temperature exposure threshold is 103; the current measured pressure is 0.872 MPa, and the normalized pressure is 0.74; the current measured flow rate is 18.0 L / min, and the normalized flow rate is 0.00, indicating that the main jet has stopped and is at the minimum stable flow rate boundary; the current average wind speed is 2.0 m / s, the angle between the wind direction and the main jet axis is approximately 60 degrees, and the normalized wind disturbance intensity is 0.58; the normalized value of the coverage change is -0.11; the normalized value of the thermal attenuation potential change is 0.09; the normalized value of the flame color response is 0.03, the normalized value of the flicker response is 0.05, and the flame visibility index is 0.04.

[0116] In step S2, the processor calculates the current foam coverage rate as 0.785 based on the foam mask, indicating that approximately 78.5% of the surface is still covered by foam. If the coverage rate in the previous cycle was 0.807, then the coverage rate has decreased. Further calculation based on the overlap between the foam edges and high-temperature pixels yields a foam layer damage index of 0.249. This result indicates that although the overall coverage rate is high, approximately one-quarter of the edges are exposed to high temperatures, making it inappropriate to determine safety solely based on the overall coverage rate.

[0117] In step S3, the processor obtains a hot spot area ratio of 0.080 based on the number of high-temperature hot spot pixels, which is normalized to 0.40; it obtains a temperature gradient energy of 10.5 degrees Celsius per pixel based on thermal image gradient statistics, which is normalized to 0.42; and it obtains a temperature rise reversal factor of 0.25 based on the average temperature change of the high-temperature region in the current and previous cycles. After fusing these three factors, the thermal decay potential index is 0.353. Since this index is higher than the thermal decay threshold of 0.35, it indicates that the current hot core has not yet entered a low-risk stable decay state.

[0118] In step S4, the processor determines that the system is currently in the heat suppression stage based on the flame visibility index (0.04) being lower than 0.12 and the heat decay potential index (0.353) being not lower than 0.35. The reignition risk index, calculated based on the weighting of the heat suppression stage, is 0.286. This result is higher than the warning threshold (0.22) but lower than the mandatory respray threshold (0.45). Therefore, the system enters the fixed-point respray preparation state instead of directly stopping respraying.

[0119] In step S5, the processor identifies two high-demand regions that meet the minimum effective area requirement based on the local highest temperature region, the proximity of the damaged edge, and the brightness attenuation distribution. The total area ratio of the high-demand regions is 0.18. Further, a supplementary spray intensity command of 0.395 is generated based on the reignition risk index and the total area ratio of the high-demand regions. For the main implementation path of the fixed nozzle, the valve opening time corresponding to this supplementary spray intensity is 2.06 s. The processor issues the command in advance after considering a 0.12 s action delay. For steerable nozzles, the processor can also map the center of gravity of the maximum demand area to the target spray azimuth, such as approximately 12 degrees to the right front and a downward adjustment of approximately 4 degrees of pitch angle.

[0120] After the respraying was completed, in the 37th control cycle, the system re-collected data and evaluated the following: coverage increased to 0.842, damage index decreased to 0.161, thermal decay potential decreased to 0.271, and reignition risk decreased to 0.173. Since this risk value was below the warning threshold and the total area of ​​high-demand areas was relatively low, the system stopped further respraying and entered observation mode.

[0121] This implementation method uses a digital fire scene simulation for verification. The total simulation duration is 180 seconds, with the first 40 seconds being the main injection phase and the following 140 seconds being the maintenance and observation period. The initial equivalent temperature of the fuel surface is 620 degrees Celsius. The normal operating wind speed is 1.2 m / s, and the impact condition is a superimposed crosswind of 3.6 m / s from 55 to 75 seconds. The control schemes include a static stop-fire scheme based on the average temperature threshold and a fixed 30-second timed supplementary injection scheme. The statistical results of this implementation method are as follows: the average warning advance before reignition is 20.8 ± 2.7 seconds, the foam consumption is 95.8 ± 5.4 L, the duration of high risk is 28.9 ± 3.7 seconds, and the reignition rate within 180 seconds is 7.2% ± 1.8%. Compared with the control scheme, this demonstrates that by introducing a coupled assessment of the foam layer breakage index and the thermal decay potential index, and directly mapping the risk to a fixed-point supplementary injection action, reignition can be suppressed earlier without simply increasing the injection volume.

[0122] Subsequently, based on the above calculation and processing results, the processor continues to execute subsequent steps in the overall process, such as task data archiving, parameter update determination, and anomaly detection. Its specific execution logic is consistent with implementation method one, and will not be repeated here.

[0123] Implementation Method 3: Phased Assessment and Abnormal Degradation Control of Warehouse Stack Surfaces under Fire Scenarios

[0124] This embodiment describes a fire scenario on the surface of a packaging material storage stack. In this scenario, the fire surface has an irregular shape, smoke obscures the surface more significantly, and the probability of sporadic frame drops in thermal imaging is relatively high. Without phased judgment criteria and an anomaly degradation mechanism, the system is prone to prematurely exiting the re-spraying when the flame is no longer visible but there are still heat cores inside the stack, or directly stopping control when a short-term anomaly is observed.

[0125] In this embodiment, the equivalent projected area of ​​the fire-covered region is 3.2 m × 2.6 m; the thermal image resolution is 64 × 64; the visible light resolution is 640 × 640; the control cycle remains 1 s; the injection pressure is 0.78 MPa; the main injection flow rate is 22 L / min; the normal environmental wind disturbance is 0.6 m / s; the continuous frame loss event in the thermal image is set to 2 to 3 frames, occurring between 88 and 90 seconds. The remaining stage thresholds, action thresholds, and degradation criteria are consistent with those in Embodiment 1 to ensure a unified parameter system.

[0126] In the 88th control cycle, the system is at the forefront of the containment phase after the open flame has been extinguished. The processor receives the following data: foam coverage is 0.812, foam layer damage index is 0.214, thermal decay potential index is 0.318, flame visibility index is 0.03, normalized ambient wind disturbance intensity is 0.22, and the thermal decay potential index of the previous cycle was 0.331, indicating a slight decrease in thermal decay potential. Based on the phase criteria, the processor identifies that it has entered the containment phase and calculates the reignition risk index as 0.196 according to the containment phase weight. Since this value does not exceed the re-spraying threshold, the system continues to be monitored.

[0127] In control cycles 89 and 90, thermal images experienced consecutive frame drops, and the visible light image was partially obscured by smoke. In step S701, the processor detected two consecutive frame drops in the thermal image and a decrease in the effective visible light pixel percentage to 48%, triggering an observation anomaly. Therefore, instead of forcibly calculating new thermal attenuation potential using the missing thermal images, the system entered degradation mode. The system inherited the most recent effective risk index of 0.196 and calculated a conservative supplementary spray intensity of 0.478, corresponding to a fixed nozzle opening duration of 2.33 s. Since the anomaly did not last more than three cycles, the system maintained conservative supplementary spray and waited for sensor recovery.

[0128] In the 91st control cycle, the thermal image returned to normal. The processor re-executed the normal assessment process, resulting in coverage increasing to 0.856, damage index decreasing to 0.143, thermal decay potential decreasing to 0.224, and reignition risk decreasing to 0.121. The system then exited the degradation mode and resumed normal dynamic assessment. This process demonstrates that inheriting the most recent effective risk index and performing conservative respraying in degradation mode can maintain basic coverage during short-term sensor blindness, avoiding premature respraying stoppage due to missing input. This implementation was verified using a digital twin environment of warehouse stack surface fire. The total simulation duration was 240 s, with an initial number of 5 high-temperature cores on the surface, and open flames were suppressed in approximately 70 s. Impact conditions included thermal image frame loss from 88 s to 90 s and local ventilation increased to 2.4 m / s from 120 s to 135 s. Comparative schemes included a scheme with unified risk threshold dynamic scoring and no degradation mechanism, and a scheme with timed respraying and manual verification. The statistical results of this embodiment are as follows: the reignition rate within 120 seconds after the open flame is 9.4% ± 2.0%, the proportion of misjudged stable state is 6.3% ± 1.7%, the successful fire control rate is 92.6% ± 2.4%, and the duration of continuous loss of control under abnormal conditions is 7.4 ± 1.5 seconds. Compared with the control scheme, this invention can maintain higher risk identification accuracy during the protection phase and maintain better control continuity under conditions of thermal imaging anomalies and smoke obstruction. In 8 stack surface fire control experiments, the control scheme without degradation mechanism experienced 2 premature spray stoppages during the thermal imaging anomaly phase, while this embodiment did not experience such spray stoppages; with a similar total foam usage, the average number of reignitions per mission decreased from 0.50 times / mission to 0.13 times / mission. This indicates that the abnormal degradation and safety fallback mechanism not only improves the robustness under abnormal conditions but also effectively reduces the control risk caused by short-term observation failures. Subsequently, based on the above calculation and processing results, the processor continues to execute the following steps in the overall process, such as online sample accumulation, incremental update determination, and version rollback determination. The specific execution logic is the same as that of Implementation Method 1, and will not be repeated here.

[0129] Implementation Method 4: Electronic Equipment

[0130] This embodiment provides an electronic device for executing the dynamic feedback evaluation method for compressed air foam fire extinguishing as described in any of the foregoing embodiments. The electronic device includes a memory, a processor, and a communication interface.

[0131] The memory is used to store computer programs, parameter configuration data, offline calibration results, control cycle cache data, and historical task data. The parameter configuration data includes at least foam mask extraction thresholds, hot spot thresholds, stage thresholds, action thresholds, equipment safety parameters, and online update version information.

[0132] The processor, connected to the memory, is used to call and execute computer programs stored in the memory to achieve the following functions: acquiring infrared thermal images, visible light images, spraying condition data, and environmental disturbance data; performing unified time alignment and spatial registration on multi-source data; constructing foam coverage integrity status and thermal attenuation potential status; performing fire extinguishing stage identification and reignition risk assessment; generating respraying areas and intensities based on reignition risk and respraying demand maps; executing degradation respraying rules in case of observational or execution anomalies; and performing parameter calibration and online update related processing after the task is completed. The communication interface is used for data interaction with the infrared thermal imager, visible light camera, pressure sensor, flow meter, wind speed and direction sensor, and spraying actuator. The communication interface can adopt a serial bus interface, industrial Ethernet interface, wireless communication interface, or other communication interface forms suitable for fire equipment linkage control.

[0133] In one alternative embodiment, the electronic device may be an industrial controller, an edge computing host, an embedded control terminal, or a dedicated processing device integrated into the control cabinet of a foam fire extinguishing system. When the processor executes the computer program, it can implement the method steps in the foregoing embodiments.

[0134] The embodiments of the present invention have been described in detail above. Those skilled in the art can make reasonable adjustments to the parameter configuration according to the application scenario, fire scale and equipment configuration without departing from the basic principles of foam fire extinguishing and heat conduction of the present invention. Such equivalent substitutions and adaptive adjustments all fall within the protection scope defined by the claims of the present invention.

Claims

1. A dynamic feedback evaluation method for compressed air foam fire extinguishing, characterized in that, include: Infrared thermal images, visible light images, spray condition data, and environmental disturbance data are collected during the compressed air foam fire extinguishing process. The collected data are then uniformly time-aligned and spatially registered to obtain multi-source fire field data corresponding to the same control cycle. Based on the multi-source fire data, foam area masks are extracted, foam coverage and its changes are calculated based on the foam area masks, and foam layer damage index is calculated based on the correspondence between the edge of the foam area mask and the high-temperature exposed area in the infrared thermal image, so as to construct a coverage state that characterizes the integrity of foam coverage. Based on the infrared thermal image, high-temperature hot spot regions are extracted, and the temperature rise reversal factor corresponding to the periodic changes of hot spot area ratio, temperature gradient energy, and average temperature of high-temperature regions is calculated. The thermal decay potential index is calculated based on the hot spot area ratio, temperature gradient energy, and temperature rise reversal factor to construct a thermal state characterizing residual heat release. Based on the visible light image, visible flame features are extracted to generate a flame visibility index. The fire extinguishing process is divided into flame control, heat suppression, and coverage stages according to the flame visibility index and the heat decay potential index. According to the current stage, the foam layer damage index, heat decay potential index, foam coverage, environmental disturbance data, and heat decay potential change are weighted to obtain the reignition risk index. Based on the reignition risk index and the replenishment demand map constructed from local thermal risk characteristics, edge damage spatial characteristics, and foam decay spatial characteristics, multiple high-demand areas are sorted by risk intensity, and the replenishment areas and replenishment intensities are determined in sequence. The compressed air foam injection mechanism is then driven to perform fixed-point closed-loop replenishment.

2. The dynamic feedback evaluation method for compressed air foam fire extinguishing according to claim 1, characterized in that, The process of performing unified time alignment and spatial registration on the collected data includes: Using the control cycle as a unified time reference, interval statistics are performed on injection condition data with a sampling frequency higher than the control cycle, and time-sequence-preserving cycle correspondence is performed on image data with a sampling frequency lower than the control cycle. The visible light image is reprojected onto the infrared thermal image coordinate system using the pre-calibrated mapping relationship, so that the foam surface image and the infrared thermal image correspond in the same spatial coordinate system. Outlier filtering is performed on continuous numerical data in pressure, flow and environmental disturbances, and neighborhood replacement is performed on saturated pixels in thermal images. When the number of consecutive missing frames does not exceed the preset number, interpolation is performed to complete the image. When the number of consecutive missing frames exceeds the preset number, an anomaly flag is set and a degradation strategy is triggered.

3. The dynamic feedback evaluation method for compressed air foam fire extinguishing according to claim 1, characterized in that, The process of extracting foam area masks based on the multi-source fire data, calculating foam coverage and its changes based on the foam area masks, and calculating the foam layer damage index based on the correspondence between the edge of the foam area mask and the high-temperature exposed area in the infrared thermogram includes: The spatially registered visible light image is converted to a color space that includes luminance and saturation components, and foam pixels are determined by combining local texture uniformity to generate a foam region mask. The foam coverage rate is calculated based on the pixel ratio of the foam area mask in the monitoring area, and the coverage rate change is generated based on the difference in foam coverage rate between adjacent control cycles. Extract the edge set of the foam region mask and count the proportion of edge pixels in the edge set that correspond to thermal image pixels that are higher than the preset high temperature exposure threshold to obtain the foam layer damage index; When the foam coverage rate is lower than the preset coverage limit, the current coverage status is directly judged as coverage failure.

4. The dynamic feedback evaluation method for compressed air foam fire extinguishing according to claim 1, characterized in that, The calculation of the thermal decay potential index based on the hot spot area ratio, temperature gradient energy, and temperature rise reversal factor includes: The hot spot area ratio is calculated based on the percentage of pixels in the infrared thermal image that are above a preset hot spot threshold. The temperature gradient energy is calculated based on the intensity of the temperature difference between adjacent pixels in the infrared thermal image in the horizontal and vertical directions. Based on the average temperature change in the high-temperature region between the current control cycle and the previous effective control cycle, a temperature rise reversal factor is generated to characterize the trend of thermonuclear recovery. The hot spot area ratio, temperature gradient energy, and temperature rise reversal factor are fused according to preset weights to obtain the thermal decay potential index, and the thermal decay potential change is generated based on the difference in thermal decay potential indices between adjacent control cycles.

5. The dynamic feedback evaluation method for compressed air foam fire extinguishing according to claim 1, characterized in that, The process involves extracting visible flame features from the visible light image, generating a flame visibility index, and dividing the fire extinguishing process into flame control, heat suppression, and coverage stages based on the flame visibility index and the heat attenuation potential index. The flame color response is generated based on the area ratio of warm-colored pixels in the visible light image; A time-domain flicker response is generated based on the degree of fluctuation in the area sequence of warm-colored regions over multiple consecutive control cycles; The flame color response and time-domain flicker response are fused according to preset weights to obtain the flame visibility index; When the flame visibility index is not lower than the preset flame visibility threshold, it is determined to be the flame control stage; when the flame visibility index is lower than the preset flame visibility threshold and the heat decay potential index is not lower than the preset heat decay threshold, it is determined to be the heat suppression stage; when the flame visibility index is lower than the preset flame visibility threshold and the heat decay potential index is lower than the preset heat decay threshold, it is determined to be the protection stage.

6. The dynamic feedback evaluation method for compressed air foam fire extinguishing according to claim 5, characterized in that, According to the current stage, the re-ignition risk index is obtained by weighting the foam layer damage index, thermal decay potential index, foam coverage, environmental disturbance data, and thermal decay potential change, including: Construct a state vector that includes at least foam coverage, foam layer damage index, thermal decay potential index, normalized injection pressure, normalized foam liquid flow rate, normalized environmental wind disturbance intensity, coverage change, and thermal decay potential change. Different risk factor weights are set for the flame control stage, heat suppression stage, and heat preservation stage; At the current stage, the re-ignition risk index is obtained by weighting and integrating the foam layer damage index, thermal decay potential index, insufficient foam coverage, environmental wind disturbance intensity, and the upward trend of thermal decay potential. The reignition risk index is compared with the stop spraying threshold, the warning threshold, and the forced spraying threshold to determine the stop spraying, the preparation for fixed-point spraying, and the high-priority spraying status.

7. The dynamic feedback evaluation method for compressed air foam fire extinguishing according to claim 1, characterized in that, The method involves using the reignition risk index and a respraying demand map constructed from local thermal risk characteristics, edge damage spatial characteristics, and foam decay spatial characteristics to sort multiple high-demand areas by risk intensity, and then sequentially determining the respraying areas and respraying intensity, including: Normalized temperature values ​​are generated based on the temperature distribution in the infrared thermal image, the proximity of the damaged edge is generated based on the distance from the pixel to the edge of the mask in the foam region, and local evaporation characterization values ​​are generated based on the brightness decay of the foam surface within a continuous control cycle. The normalized temperature value, the proximity of the damaged edge, and the local evaporation characterization value are fused according to preset weights to generate a pixel-level respray demand map. Extract high-demand connected regions from the respray demand map, filter them by area to obtain a set of respray regions, and calculate the total area ratio of high-demand regions. A supplementary spraying intensity command is generated based on the reignition risk index and the total area ratio of the high-demand area.

8. The dynamic feedback evaluation method for compressed air foam fire extinguishing according to claim 7, characterized in that, The driving compressed air foam injection mechanism performs fixed-point closed-loop supplementary spraying, including: When the reignition risk index is lower than the stop spraying threshold and the total area of ​​high-demand areas is relatively low, a stop spraying command is output. The supplementary spray intensity command is mapped to a supplementary spray control quantity corresponding to the type of spray mechanism; Among them, the valve opening time corresponding to the fixed nozzle is compensated by the action delay of the actuator before the control is issued; the center of gravity of the supplementary spray area is mapped to the nozzle target direction for the directional nozzle, and the supplementary spray intensity command is mapped to the corresponding spray action before the control is issued. In the next control cycle after the supplementary spraying action is executed, data is collected again and the evaluation and supplementary spraying decision are repeated to form a closed loop.

9. The dynamic feedback evaluation method for compressed air foam fire extinguishing according to claim 1, characterized in that, It also includes parameter calibration and online update steps, which include: An offline sample set was constructed based on physical simulation samples and controlled experimental samples, labeled with whether the fire would reignite within a future preset observation period; A joint objective function, including the losses from reignition underreporting, overspraying, and identification lag, is used to calibrate the stage threshold, stage risk weight, and fusion weight in the spraying demand graph offline. When the accumulated online task data reaches the preset quantity, only the stage threshold, stage risk weight and fusion weight in the replenishment demand graph are incrementally updated, while the device safety parameters remain frozen. If the re-ignition false alarm rate increases or the invalid re-spray rate fails to meet the preset improvement conditions in the updated version, roll back to the previous stable version.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the dynamic feedback evaluation method for compressed air foam fire extinguishing as described in any one of claims 1 to 9.

Images