FPSO intelligent directional spraying method based on multi-sensor fusion

Abstract

The invention discloses an FPSO (Floating Production Storage and Offloading) intelligent directional spraying method based on multi-sensor fusion, which is suitable for fire prevention and control of an FPSO upper module. The method comprises the following steps: firstly, dividing protection sub-areas, arranging infrared, ultraviolet, visible light and attitude sensors, matching with mixed nozzles and networking; establishing a module coordinate system, and completing sensor calibration and multi-modal data registration; and a multi-task CNN fire identification model is constructed, and fire source positioning, type discrimination and spreading prediction are realized. When the system operates, image stabilization is performed through hull attitude compensation, fusion feature input models are extracted, and personalized fire extinguishing strategies are generated; and generating a spraying instruction through space calculation and the like, driving the nozzle to spray in a directional and constant-flow manner, and dynamically optimizing the strategy based on real-time feedback until the fire is extinguished. Point-to-point fire extinguishing is achieved, water stain loss and fire fighting water waste are reduced, and the accuracy and efficiency of fire fighting in the marine environment are improved.

Description

Technical Field

[0001] This invention relates to fire-fighting methods for marine engineering, and more particularly to an intelligent directional spraying method for the top module of a floating production storage and offloading (FPSO) unit, based on multi-sensor fusion and convolutional neural network (CNN) recognition. Background Technology

[0002] The FPSO (Floating Production Station) integral floating structure, or hull, is the core facility for offshore oil and gas development. Its superstructure integrates a large number of complex process equipment and electrical instrumentation. The FPSO superstructure refers to functional units fixed to the main deck of the hull, such as process areas, power generation modules, and electrical rooms. Its structure is a multi-layered steel frame system. The internal structure of the FPSO superstructure includes the ceiling, equipment deck, main load-bearing beam and column system, and external enclosure structure. The modules are rigidly connected to the main deck 3-1 of the hull by welding or high-strength bolts. The ceiling 3-3 is located at the top of the module, typically the lower surface of the module's upper deck or an independently hoisted fireproof and explosion-proof panel, used to install sprinkler system heads, cable trays, lighting, and ventilation ducts, among other auxiliary facilities. The space below it houses various process equipment, pipelines, and control systems. Fires in the FPSO superstructure pose a high risk and have serious consequences.

[0003] Currently, the traditional sprinkler or water mist fire suppression systems commonly used in FPSO upper modules are mostly designed for "full area coverage." Once a fire detector alarms, the system sprays the entire protected area. This approach has significant drawbacks: First, the large amount of spray water can cause serious secondary damage to undisturbed precision electrical equipment and control systems, leading to huge economic losses and lengthy recovery periods. Second, full-coverage spraying results in a significant waste of fire-fighting water (especially fresh water), which is particularly disadvantageous for offshore FPSOs with limited resupply capabilities. Finally, traditional systems cannot differentiate between fire types and scales, and cannot implement differentiated optimal fire suppression strategies, thus their fire suppression efficiency needs improvement.

[0004] In recent years, intelligent fire protection has developed in the fields of land construction and industry. However, its application is usually based on relatively single or few sensors (such as smoke and temperature). In the complex and dynamic marine engineering environment, with dense equipment, diverse fire types (such as pool fire, jet fire, and electrical fire), and extreme conditions accompanied by continuous swaying of the ship, existing technologies lack effective multi-source information fusion processing capabilities, high-precision fire identification capabilities, and adaptive precision execution capabilities. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide an intelligent directional spraying method for FPSOs based on multi-sensor fusion. This method can identify the location, type, and scale of fire sources in real time and accurately, and drive the execution unit to perform directional and appropriate spraying for fire extinguishing, thereby minimizing water damage, protecting precision equipment, and improving fire extinguishing efficiency and safety.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: The present invention provides an intelligent directional spraying method for FPSO based on multi-sensor fusion, comprising the following steps: Step 1: Construct a protection sub-zone and integrate a multi-source fusion monitoring terminal within the FPSO upper module. Arrange the monitoring terminal and nozzles within the FPSO upper module and network them using an industrial wireless network. Install a solenoid valve and an electrically controlled flow regulating valve on the pipeline branch of each protection sub-zone. The monitoring terminal includes a fire detection sensor and an attitude sensor. The fire detection sensor includes an infrared thermal imager, an ultraviolet flame detector, and a visible light camera. Within each zone, a hybrid nozzle layout is adopted, including adjustable nozzles and fixed nozzles. The adjustable nozzles are connected to a rotary drive device. The drive controller of the rotary drive device, the solenoid valve of the protection sub-zone, and the electrically controlled flow regulating valve are collectively referred to as the execution unit. Step 2: Establish a module coordinate system and build a model of the FPSO hull in the edge computing server; Step 3: Perform spatial pose and detection parameter calibration on the three types of fire detection sensors: infrared thermal imager, visible light camera, and ultraviolet flame detector. The calibration yields the intrinsic and extrinsic parameter matrices of the infrared thermal imager and visible light camera, as well as the effective response area parameters of the ultraviolet detector. After calibration, the infrared images acquired by the infrared thermal imager and the visible light images acquired by the visible light imager are unified to the module coordinate system. At the same time, the effective response area of ​​the ultraviolet detector obtained from the calibration is unified to the module coordinate system through coordinate transformation to form spatial registration data, which is then stored in the edge computing server. Step 4: Based on the spatial registration data from Step 3, map the existing multimodal fire data to the module coordinate system. Use the mapped multimodal fire data to construct and train a convolutional neural network fire identification model that fuses multi-source information. The convolutional neural network fire identification model outputs structured results: three-dimensional spatial coordinates of the fire source, fire type, and fire spread direction vector. Step 5: After the system enters the running state, the edge computing server uses the ship attitude data collected by the attitude sensor to perform image stabilization processing on the infrared images collected by the infrared thermal imager and the visible light images collected by the visible light camera; using the calibration parameters obtained in Step 3, the image-stabilized infrared images and visible light images are input into the module coordinate system; the edge computing server extracts fire-related features from the image-stabilized infrared images, visible light images and the spatially registered ultraviolet flame detection signals, and constructs a multi-channel fusion feature tensor. A four-channel fusion feature tensor is constructed and input into the trained convolutional neural network fire recognition model for real-time processing to obtain the fire situation recognition results. Finally, the intelligent decision-making module deployed on the edge computing server generates the corresponding fire extinguishing strategy based on the fire situation recognition results. Step Six: Based on the fire extinguishing strategy in Step Five, the edge computing server calculates the number of adjustable sprinkler heads to be activated to obtain the corresponding sprinkler control command. The sprinkler control command includes at least: fire type, activated adjustable sprinkler head number and fixed sprinkler head number, sprinkler mode, three-dimensional target coordinates of the fire source in the module coordinate system, and sprinkler flow rate level. Then, it is sent to the corresponding execution unit through the industrial wireless network in a time sequence. The execution unit performs the sprinkler fire extinguishing operation according to the fire extinguishing strategy and executes the sprinkler control command. After the execution unit starts execution, the intelligent decision module enters the feedback monitoring state and continuously receives real-time data of the fire source area collected by the infrared thermal imager, ultraviolet flame detector and visible light camera. Step 7: The intelligent decision-making module conducts dynamic evaluation and adaptive optimization control of the fire extinguishing strategy execution effect based on real-time fire source monitoring data under feedback monitoring status.

[0007] The beneficial effects of this invention are: 1. Precise fire suppression, reduced water consumption. Through multi-sensor fusion and CNN recognition, fire sources are located and classified, transforming "area coverage" into "point-to-point fire suppression," greatly reducing fire water consumption and secondary damage to undisturbed precision equipment. 2. Intelligent decision-making, improved efficiency. The system automatically adopts the optimal fire suppression strategy for different fire types (pool fire, jet fire, electrical fire), improving the initial fire suppression success rate and effectively controlling the fire scale and losses.

[0008] 2. High adaptability and stability: The marine environmental data processing algorithm of this invention effectively reduces the impact of harsh working conditions such as ship swaying and salt spray corrosion on the system's sensing capabilities, ensuring the long-term stable operation and high reliability of the system.

[0009] 3. Closed-loop control and dynamic optimization: It has the ability to provide real-time feedback and adjust strategies, and can dynamically optimize the sprinkler scheme according to changes in the fire situation to achieve optimal control of the fire extinguishing process.

[0010] 4. This invention enables rapid, accurate, and adaptive fire suppression of fires on the upper modules of FPSOs, maximizing asset protection while significantly improving the intelligence level and resource utilization efficiency of the fire protection system. Attached Figure Description

[0011] Figure 1 This is a schematic diagram of the overall architecture and information flow of the method of the present invention; Figure 2 This is a schematic diagram illustrating the implementation of personalized sprinkler strategies for different fire types.

[0012] Figure 3 This is a schematic diagram of the upper module structure of an FPSO. Figure 4 This is a schematic diagram of the nozzle drive structure. Detailed Implementation

[0013] The present invention will now be described in detail with reference to the accompanying drawings: The method of this invention is applicable to intelligent directional spray fire suppression when a fire occurs on the hull of an FPSO.

[0014] Based on the above structure, the present invention provides an intelligent directional spraying method for FPSO based on multi-sensor fusion, comprising the following steps: Step 1: Construct a protection sub-zone and integrate a multi-source fusion monitoring terminal within the FPSO upper module. Arrange the monitoring terminal and nozzles within the FPSO upper module and establish an industrial wireless network. The monitoring terminal includes a fire detection sensor and an attitude sensor. The fire detection sensor includes an infrared thermal imager, an ultraviolet flame detector, and a visible light camera. Specific steps are as follows: Step 101: Fire compartmentation and sprinkler arrangement are carried out in the protected space to form a fire suppression capability that combines basic coverage with precise fire suppression; the details are as follows: The protected area of ​​the FPSO upper module is divided into multiple independent protected sub-zones, and a fire water supply network is laid out within the structural mezzanine of the module ceiling (3-3). In this step, the protected area is divided into multiple independent protected sub-zones, employing a zoned control design. A solenoid valve and an electrically controlled flow regulating valve are installed on the pipe branches of each protected sub-zone to regulate the water supply flow rate of that sub-zone. The area of ​​each protected sub-zone is determined based on the fire risk level of its internal equipment; a recommended embodiment is 10m × 10m to 15m × 15m.

[0015] Preferably, the low, medium, and high speeds of the electronically controlled flow regulating valve correspond to 20%-40%, 60%-80%, and 95%-100% of the rated flow rate Q0 of a single adjustable nozzle, respectively. The discretized design of the flow rate levels is based on the joint optimization results of the nonlinear curve of fire extinguishing efficiency and the valve control characteristics. Below 20% is the ineffective spray zone; 40%-60% is the efficiency ambiguity zone for pool fires and electrical fires, which is neither economical nor safe; 80%-95% is the zone of marginal flow reduction, where increasing the water volume has a negligible contribution to fire extinguishing. Locking the levels within these three optimal efficiency ranges ensures fire extinguishing reliability while avoiding inefficient and unstable zones.

[0016] Within each protected sub-zone, a hybrid nozzle layout is adopted, including adjustable nozzles and fixed nozzles. The adjustable nozzles are connected to a rotary drive device. The drive controller of the rotary drive device, the solenoid valve of the protected sub-zone, and the electrically controlled flow regulating valve are collectively referred to as the execution unit.

[0017] Specifically: Adjustable sprinkler head: One adjustable sprinkler head is arranged at the center of each protected sub-zone or directly above critical equipment (referring to equipment that may cause a fire); the rotary drive device is a direct-drive dual motor, which includes a horizontal rotary stepper motor 4-1 mounted on the ceiling 3-3, with the output shaft of the horizontal rotary stepper motor set vertically downward; the body of the vertical rotary stepper motor 4-2 is fixedly connected to the output shaft of the horizontal rotary stepper motor, with the output shaft of the vertical rotary stepper motor set in the horizontal direction; the adjustable sprinkler head 4-3 is fixedly connected to the output shaft of the vertical rotary stepper motor.

[0018] Around the adjustable nozzles, multiple fixed nozzles are evenly arranged in a grid pattern according to the water spray intensity and maximum spacing required by the "Safety Code for Fixed Offshore Platforms" or relevant classification society regulations. These fixed nozzles are installed on ceiling 3-3. The spray angles of these fixed nozzles are pre-adjusted and fixed according to their positions during installation to cover any potential blind spots of the adjustable nozzles within this sub-area, complementing the coverage area of ​​the adjustable nozzles.

[0019] More preferably, the adjustable nozzle and the fixed nozzle are explosion-proof nozzles.

[0020] In this structure, when the horizontal rotary stepper motor 4-1 is driven, it directly drives the vertical rotary motor 4-2 and the adjustable nozzle 4-3 to rotate as a whole in the horizontal plane; when the vertical rotary stepper motor is driven, it directly drives the nozzle to rotate vertically around the horizontal axis. This dual-motor direct-drive architecture realizes direct and independent control of the two degrees of freedom, horizontal and vertical rotation. Its dual-axis integrated positioning accuracy (i.e., the maximum angular deviation between the central axis of the nozzle and its theoretically calculated target pointing line) is guaranteed by the motor control accuracy and can reach ≤0.5°.

[0021] Step 102: Distribute and install monitoring terminals to obtain real-time fire and ship attitude information; The fire detection sensors are installed inside the FPSO upper module, on rigid supports 3 to 5 meters above the main deck 3-1 (i.e., the hull deck plane supporting the upper module), and on the equipment deck 3-4. The installation position ensures that their combined field of view completely covers all protected equipment areas downwards. As an example, the infrared thermal imager has a temperature measurement range of -20℃ to 600℃, the ultraviolet flame detector has a response wavelength of 185nm to 245nm, and the visible light camera has a frame rate of 30 frames per second.

[0022] The attitude sensor is used to measure the ship's roll angle in real time. and pitch angle The attitude sensor is rigidly mounted on the main steel structure 3-2 (which can be the main load-bearing beam) of the upper module of the FPSO. In one embodiment, the attitude sensor can be a nine-axis attitude sensor (IMU).

[0023] All nozzles and sensors move together with their mounting bases and the hull, so subsequent motion compensation for fire source positioning is required using data measured by the attitude sensors.

[0024] Step 103: Connect all monitoring terminals and execution units to the edge computing server located in the secure area via the industrial wireless network to form a system control loop.

[0025] The edge computing server is used to receive raw or pre-processed data from infrared thermal imagers, ultraviolet flame detectors, visible light cameras, and attitude sensors, and to provide a unified data processing and computing platform for subsequent fire source spatial positioning, motion compensation calculation, and sprinkler control decisions.

[0026] The edge computing server is preferably an explosion-proof edge computing server.

[0027] Step 2: Establish a module coordinate system and a model of the FPSO hull in the edge computing server. The FPSO hull model includes a model of the equipment layout of the upper FPSO module and the calibration of the installation coordinates of the monitoring terminal and the nozzle. The specific method is as follows: When the FPSO hull is in its initial horizontal state, select a physical feature point inside the upper module of the FPSO as the module coordinate system O. m -X m Y m Z m Origin m The physical feature point can preferably be a fixed, easily measurable, and less susceptible to obstruction or collision by equipment. The axes of the module coordinate system are defined as follows: X m The axis points towards the bow, Y m The axis points to the starboard side, Z m The axis points vertically upwards. The unified modular coordinate system established in this step provides a unique static geometric reference for the joint calibration and spatial registration of subsequent fire detection sensors. All calibration results will be mapped to this coordinate system.

[0028] Preferably, the physical feature point can be a fixed location that is easy to measure and not easily obstructed or collided with by equipment. More preferably, the physical feature point can be selected from the center of a specific bolt hole on the structural column of the main deck 3-1 of the hull, or a welding point of the main structure of the module confirmed by surveying.

[0029] Step 3: The spatial pose and detection parameters of the three types of fire detection sensors—infrared thermal imager, visible light camera, and ultraviolet flame detector—are calibrated separately. The calibration yields the intrinsic and extrinsic parameter matrices of the infrared thermal imager and visible light camera, as well as the effective response area parameters of the ultraviolet detector. After calibration, the infrared images acquired by the infrared thermal imager and the visible light images acquired by the visible light imager are unified to the module coordinate system. Simultaneously, the effective response area of ​​the ultraviolet detector obtained from the calibration is unified to the module coordinate system through coordinate transformation, forming spatial registration data, which is stored in the edge computing server. The specific implementation is as follows: Step 301, Visual sensor calibration data acquisition: The Zhang Zhengyou planar calibration method (see Zhang, Z.(2000). A flexible new technique for camera calibration. IEEE Transactions on Pattern Analysis and Machine Intelligence, 22(11), 1330-1334.) is used to calibrate the infrared thermal imager and the visible light camera (excluding the ultraviolet flame detector) respectively. Specifically, multiple images of the standard calibration board in different postures are captured by the infrared thermal imager and the visible light imaging head respectively to obtain image data for calibration. Step 302, Camera Parameter Solving: Based on the acquired standard calibration board image, establish the geometric mapping relationship between pixels and three-dimensional spatial points (e.g., calibration board corner points) in the image pixel coordinate system. Any pixel in the image pixel coordinate system is represented as: p = [u, v] Where u represents the horizontal coordinate of a pixel and v represents the vertical coordinate of a pixel.

[0030] The calibration parameters of the infrared thermal imager and the visible light camera are calculated using a calibration algorithm and then input into the edge computing server.

[0031] The calibration parameters of the infrared thermal imager and the visible light camera include the intrinsic parameter matrix K of the infrared thermal imager and the visible light camera, which includes focal length parameters, principal point coordinates and distortion parameters; and the extrinsic parameter matrices of the infrared thermal imager and the visible light camera relative to the module coordinate system, which include rotation matrix R and translation vector T, used to describe the spatial position and attitude of the infrared thermal imager and the visible light camera in the module coordinate system.

[0032] Step 303: Calibrate the effective response area of ​​the ultraviolet detector in the module coordinate system and store it in the edge computing server.

[0033] Specifically, the effective response area of ​​each ultraviolet flame detector in the module coordinate system can be determined through calibration experiments. The effective response area of ​​the ultraviolet detector is stored in the edge computing server in the form of spatial boundary points or a mathematical expression. This calibration experiment adopts the spatial calibration principle of heterogeneous sensors based on a reference target. The specific process is as follows: a movable standard flame source with known fire intensity is used as a reference target, and its position is systematically changed within the protected area. For each position, the three-dimensional coordinates of the flame source are determined using a calibrated visible light camera, and the response state of the ultraviolet detector is recorded simultaneously. Based on the collected data points, a three-dimensional spatial bounding box fitting algorithm is used to fit the effective response area of ​​the ultraviolet detector in the module coordinate system, and the data is input into the edge computing server to complete the on-site mapping of its detection area.

[0034] Step 304, Spatial Registration of Fire Detection Sensors: Based on the intrinsic and extrinsic parameter matrices of the infrared thermal imager and visible light camera obtained through calibration, the infrared images acquired by the infrared thermal imager and the visible light images acquired by the visible light imager are unified to the module coordinate system through coordinate transformation and image reprojection technology. Simultaneously, the effective response area of ​​the ultraviolet detector obtained in step 303 is unified to the module coordinate system through coordinate transformation. Thus, spatial registration of infrared, visible light, and ultraviolet multimodal sensing data is achieved. The intrinsic and extrinsic parameter matrices of the infrared thermal imager and visible light camera obtained through calibration and registration, as well as the effective response area parameters of the ultraviolet detector in the module coordinate system and the unified coordinate mapping relationship, together constitute the spatial registration data, all of which are stored in the edge computing server. The effective response area of ​​the ultraviolet detector in the module coordinate system serves as the spatial registration reference for the ultraviolet flame detection signal.

[0035] Step 4: Based on the spatial registration data from Step 3, map the existing multimodal fire data to the module coordinate system. Use the mapped multimodal fire data to construct and train a convolutional neural network fire identification model that fuses multi-source information. The convolutional neural network fire identification model outputs structured results: three-dimensional spatial coordinates of the fire source, fire type, and fire spread direction vector. The specific implementation is as follows: Step 401, Acquisition and Preprocessing of Multimodal Fire Data: Download multimodal fire data of FPSO from public marine engineering professional databases, perform standardized preprocessing on the raw data, and remove invalid and abnormal data; based on the sensor intrinsic and extrinsic parameter matrices and the spatial registration results of the ultraviolet detector obtained in Step 3, convert the preprocessed infrared, visible light and ultraviolet data into spatial data in the module coordinate system.

[0036] Step 402: Construct an annotated FPSO fire multimodal sample dataset: Annotate the spatially registered multimodal data, annotating the three-dimensional spatial coordinates of the fire source in the module coordinate system, the fire type, and the fire spread direction vector.

[0037] Step 403, Construction, Training, and Deployment of the Convolutional Neural Network Fire Recognition Model: Based on a multi-task learning framework, a convolutional neural network fire recognition model with "four-channel parallel multi-branch feature extraction and cross-channel attention fusion" is built. The model sets four parallel feature extraction branches, corresponding to infrared temperature features, visible light color features, motion features, and ultraviolet flame feature channels, respectively. The model output is a structured result corresponding to the labeled content: three-dimensional spatial coordinates of the fire source, fire type, and fire spread direction vector.

[0038] After construction, an uncertainty-weighted multi-task loss function is used to train and validate the convolutional neural network fire identification model using a labeled FPSO fire multimodal sample dataset. During training, to improve computational efficiency and inference performance, a quantum neural network (QNN) module can be optionally introduced into the convolutional neural network fire identification model to compress intermediate features or optimize the decision-making process. After training, the model parameters are deployed to an edge computing server for real-time access during system operation.

[0039] Step 5: After the system enters the running state, the edge computing server uses the ship attitude data collected by the attitude sensor to stabilize the infrared images collected by the infrared thermal imager and the visible light images collected by the visible light camera. Using the calibration parameters obtained in Step 3, the stabilized infrared and visible light images are input into the module coordinate system. The edge computing server extracts fire-related features from the stabilized infrared and visible light images and the spatially registered ultraviolet flame detection signals, and constructs a multi-channel fusion feature tensor. A four-channel fusion feature tensor is then constructed and input into the trained convolutional neural network fire recognition model for real-time processing to obtain the fire situation recognition results. Finally, the intelligent decision-making module deployed on the edge computing server generates corresponding fire extinguishing strategies based on the fire situation recognition results, including the following steps: Step 501, Monitoring terminal data synchronous acquisition and image stabilization preprocessing of visual sensor output images: In the system running state, data from the infrared thermal imager, visible light camera, ultraviolet flame detector and attitude sensor are synchronously acquired at a preset time period (e.g., 100ms), and the data from the infrared thermal imager and visible light camera are aligned with the timestamps before being transmitted to the edge computing server.

[0040] First, the ship's attitude rotation matrix is ​​calculated: the ship's roll angle is obtained in real time using attitude sensors (IMU). and pitch angle Construct the basic rotation matrices of the FPSO hull about the X-axis and about the Y-axis respectively:

[0041] And calculate the hull attitude rotation matrix at the k-th sampling time based on the above basic rotation matrix: ) The ship's roll angle is obtained in real time by the attitude sensor (IMU). The pitch angle of the hull is obtained in real time by the attitude sensor (IMU). : The ship's attitude rotation matrix at the k-th sampling time. Rotation around the X-axis The basic rotation matrix of angles, Rotation around the Y-axis The basic rotation matrix of angles.

[0042] Subsequently, using Digital image stabilization is performed on infrared and visible light images. For any pixel point p = [u, v] in the infrared and visible light images... Combining the calibration intrinsic parameter matrix K and extrinsic parameter matrix [R|T] obtained in step three, they are mapped to a stable module coordinate system through the following coordinate transformation relationship:

[0043] Among them, P m This represents the spatial point or direction vector corresponding to a pixel in the module coordinate system.

[0044] After completing the above coordinate transformation, the spatial point corresponding to any pixel in the infrared image and visible light image in the module coordinate system is reprojected onto the virtual horizontal image plane (this plane is a stationary reference plane parallel to the initial horizontal main deck 3-1 plane in the module coordinate system, and its attitude does not change with the sway of the ship), so as to obtain the stabilized infrared image and visible light image, which are stored in the edge computing server, thereby eliminating the pixel offset caused by the sway of the ship.

[0045] Step 502: Based on the image stabilization process, the edge computing server extracts fire-related features from the stabilized infrared image, visible light image, and spatially registered ultraviolet flame detection signal, and constructs a multi-channel fusion feature tensor as input for the subsequent fire identification model. The specific steps are as follows: Infrared temperature feature map construction, including: Based on the stabilized infrared image acquired by the infrared thermal imager, pixel-by-pixel temperature information is extracted and normalized to obtain an infrared temperature feature map. This infrared temperature feature map is used to characterize the high-temperature distribution characteristics of suspected fire source areas. The calculation formula is as follows:

[0046] in This represents the actual temperature value corresponding to pixel (i,j) in the infrared image. , In this embodiment, the preset temperature measurement range is set to -20℃ to 600℃. Construction of Visible Light Flame Color Feature Map: The stabilized visible light image acquired by the visible light camera is converted from the RGB color space to the HSV color space, and a flame color likelihood map is constructed based on typical flame color features. The visible light flame color feature map and motion feature map are generated respectively. The visible light flame color feature map is used to reflect the distribution of regions with flame color features in the visible light image.

[0047]

[0048] Where H(i,j), S(i,j), and V(i,j) are the hue, saturation, and luminance components of pixel (i,j), respectively. I{·} is an indicator function that takes the value 1 if the condition is met, and 0 otherwise. Motion feature map construction: Motion feature maps are extracted from stabilized visible light images acquired by a visible light camera using the inter-frame difference method. This motion feature map is used to describe the dynamic characteristics of a flame as it changes over time. The calculation method is as follows:

[0049] in: and These represent the grayscale images of the current frame and the previous frame, respectively. τ m The specific value of the motion detection threshold needs to be calibrated and determined during the system debugging phase based on the on-site environment, so as to effectively distinguish between normal disturbances and actual fire motion.

[0050] Construction of ultraviolet flame feature map: In order to realize the spatial fusion of ultraviolet detector alarm information and image data, the effective response area of ​​each ultraviolet detector has been determined in the system calibration stage (step 3).

[0051] During real-time operation, the system generates a binary ultraviolet flame feature map (UVmap(i,j)) based on the alarm status of the ultraviolet detector. This binary ultraviolet flame feature map visually represents the ultraviolet detector's judgment of the flame's presence and spatial location. Its generation rules are as follows: When the ultraviolet detector does not alarm, all pixel values ​​in UVmap(i,j) are set to 0.

[0052] When the ultraviolet detector alarms, all pixels (i, j) in UVmap(i,j) within its effective response area are assigned a value of 1, and pixels in the remaining areas are assigned a value of 0.

[0053] Multi-channel fusion feature tensor construction: The infrared temperature feature map obtained above is used to construct the multi-channel fusion feature tensor. Visible light flame color characteristic diagram Motion feature map The ultraviolet flame feature maps UVmap(i,j) are spatially aligned and then stacked along their channel dimensions to form a four-channel fused input feature tensor Xin. :

[0054] H: The height of the infrared temperature feature map, i.e., the number of pixels in the row direction of the image (Height); W: The width of the infrared temperature feature map, i.e., the number of pixels in the column direction of the image (Width).

[0055] Step 503, Fire identification and fire situation analysis based on a convolutional neural network fire identification model: The multi-channel fusion feature tensor Xin obtained in step 502 is used The data is input into a pre-trained and loaded convolutional neural network (CNN) fire identification model to perform fire identification and situation analysis.

[0056] The convolutional neural network fire identification model trained through the above process, after forward propagation calculation in the application phase, directly outputs a structured fire situation identification result through decoding at the terminal fully connected layer. This result includes at least: Fire source location information, the three-dimensional spatial coordinates of the fire source in the module coordinate system ( , , ).

[0057] Fire attribute information, fire type identification results (such as "jet fire") and their confidence level (such as 98%).

[0058] Fire dynamics information, main direction vector of fire spread ( , , ).

[0059] Output example: Fire Source ID: 01, Fire Source 3D Spatial Coordinates: ( , , Fire type: jet fire (confidence level: 98%), fire spread direction vector: ( , , )} The above fire situation identification results provide a basis for decision-making in generating subsequent personalized fire extinguishing strategies.

[0060] Step 504: Generation of fire extinguishing strategy: The intelligent decision-making module deployed on the edge computing server receives the fire situation identification result output in step 503. This result includes at least fire source location information, fire attribute information, and fire dynamic information. The intelligent decision-making module integrates a fire extinguishing rule base (such as...). Figure 2 As shown in the diagram, the intelligent decision-making module, based on the fire situation identification results, generates a fire extinguishing strategy for the current fire situation by performing rule matching and reasoning on the fire extinguishing rule base. The fire extinguishing rule base pre-stores fire extinguishing strategy rules corresponding to different fire source location information, fire attribute information, and fire dynamic information (fire type, fire source location, and fire characteristics). The fire extinguishing rule base includes at least the following strategies: For pool fires: Activate the 2-3 adjustable nozzles closest to the fire source and adopt a medium-speed scanning spray mode—control the nozzles to perform horizontal reciprocating scanning motion at a constant speed under the calculated vertical rotation angle, and at the same time set the electronically controlled flow regulating valve of the protection sub-zone to "medium" to cover and suppress the burning liquid surface area; For jet fire: Activate 1-2 adjustable nozzles upstream of the fire source and use the focused water jet mode—control the nozzles to maintain a fixed angle and accurately aim at the root of the flame, while setting the electronically controlled flow regulating valve of the protected sub-zone to "high" for cooling and shut-off; For electrical fires: The system will trigger the power supply to the equipment in the affected area. After confirming that the power outage conditions are met, the adjustable sprinkler head above the fire source will be activated in fine water mist mode. The adjustable sprinkler head will be controlled to slowly and subtly oscillate to cover the protected equipment. At the same time, the electrically controlled flow regulating valve of the protected sub-zone will be set to "low" to cool down and extinguish the fire, avoiding secondary hazards caused by high-flow water impact.

[0061] The medium-speed scanning spray mode, focused water column mode, and fine water mist mode are three spray modes of the adjustable nozzle.

[0062] Step Six: Based on the fire suppression strategy in Step Five, the edge computing server obtains the corresponding sprinkler control commands through spatial calculations of the fire source location and sprinkler installation coordinates, matching calculations of fire type and sprinkler flow rate, and optimization calculations of the number of adjustable sprinklers to be activated. These commands are then sequentially distributed to the corresponding execution units via the industrial wireless network. The execution units then perform the sprinkler fire suppression operation according to the sprinkler control commands. After the execution units begin execution, the intelligent decision-making module enters a feedback monitoring state, continuously receiving real-time fire source monitoring data collected by the infrared thermal imager, ultraviolet flame detector, and visible light camera. The specific process is as follows: Step 601, Core calculation of sprinkler control commands: Spatial calculation of fire source location and nozzle installation coordinates: calling all adjustable nozzle installation coordinates stored in the FPSO hull model ( , , () P =1, 2, ..., K, where K is the total number of adjustable nozzles), and calculate the three-dimensional spatial coordinates of each adjustable nozzle and the fire source based on the Euclidean distance formula. X 1 , Y 1 , Z 1 The spatial distance is used to select the N nearest adjustable nozzles (N is determined by the fire suppression strategy) as the target adjustable nozzles to be activated. The number of N can be as follows: for pool fires: N=2~3; for jet fires: N=1~2; for electrical fires: N=1. The horizontal rotation angle of the target adjustable nozzle is also considered. : The vertical rotation angle of the target adjustable nozzle. Values ​​range from -90° to 90°: .

[0063] Matching calculation of fire type and sprinkler flow rate: Based on the preset flow rate level of the fire extinguishing strategy (pool fire: medium level, jet fire: high level, electrical fire: low level), combined with infrared temperature characteristic map. The average temperature of the flame root region is extracted as the criterion for judgment, denoted as... ;like >400℃ (high-temperature fire), increase the strategy level by 1 level; if 200℃ ≤ ≤400℃ (normal fire situation), execute according to the strategy level; if For temperatures below 200℃ (low temperature, low heat), the original strategy settings will be uniformly lowered to the low setting.

[0064] Optimization calculation of the number of adjustable nozzles to be activated: based on the fire spread direction vector in the fire suppression strategy. ( , , ),in , , These represent the components of the fire spread direction vector along the X, Y, and Z axes of the module coordinate system, combined with the average temperature at the flame root. Select the adjustable nozzles that need to be activated. If >400℃, a maximum of 3 target adjustable nozzles can be activated; if 200℃≤ At ≤400℃, a maximum of two target-adjustable nozzles can be activated; <200℃, only one target adjustable nozzle is activated. The adjustable nozzle installation coordinates from the FPSO hull model in step two are used. , , First, select adjustable nozzles that are ≤8 meters away from the fire source in a straight line, and then further select them according to direction: When =0, select the adjustable nozzle that is closest to the fire source in the X-axis direction; When =0, select the nozzle that is closest to the fire source in the Y-axis direction; , When none of the values ​​are equal to 0, select the adjustable sprinkler closest to the main direction of fire spread and to both sides. If the fire source is ≤2 meters from the boundary of the protected sub-zone, activate one additional adjustable sprinkler from the adjacent sub-zone to eliminate the blind spot.

[0065] The standardized sprinkler control instructions generated through the above calculations include at least: fire type, target adjustable sprinkler head number and fixed sprinkler head number, sprinkler mode, and three-dimensional target coordinates of the fire source (X). m ,Y m Z m ), Horizontal rotation angle of the target adjustable nozzle Vertical rotation angle of the target adjustable nozzle Spray flow rate settings.

[0066] Step 602, Instruction Issuance and Spraying Operation Execution: The edge computing server sends sprinkler control commands to the drive controllers and electronically controlled flow regulating valves of the protection sub-zones corresponding to the adjustable sprinkler heads.

[0067] After receiving the instructions from the edge computing server, the drive controller determines the spray pattern and horizontal rotation angle based on the instructions. and vertical rotation angle The motion control algorithm pre-stored in the drive controller, corresponding to the spray mode, is invoked to complete the pre-adjustment of the nozzle attitude: In the focused water column mode, the horizontal stepper motor and the vertical stepper motor are driven to rotate horizontally by an angle. Vertical turning angle Precise rotation ensures the nozzle axis stably points towards the target point; in medium-speed scanning mode, the pre-stored trajectory planning algorithm in the drive controller is invoked to fix the vertical angle of the adjustable nozzle. The horizontal angle of the adjustable nozzle can be adjusted to... Using the baseline value, within the preset scan amplitude Δ A continuous reciprocating scanning trajectory is generated within a range of ±20°. A uniform nozzle oscillation is achieved using an interpolation algorithm; in fine water mist mode, a pre-stored micro-motion control algorithm within the drive controller is invoked, using the horizontal rotation angle obtained from the aforementioned inverse kinematics calculation. Vertical turning angle Using the angle reference, a tiny periodic oscillation trajectory (such as a sine wave) is generated, causing the nozzle to... and Small, slow movements nearby create a fine mist.

[0068] The electronically controlled flow regulating valve receives the flow level instruction from the edge computing server and adjusts the valve opening to near the target level. After the nozzle posture and flow rate are adjusted to the correct positions, the edge computing server sends the spray control command to the solenoid valve of the protected sub-zone, and the solenoid valve performs the corresponding on / off control.

[0069] After the execution unit completes the above actions, the intelligent decision-making module enters the feedback monitoring state, that is, it continuously receives real-time data of the fire source area collected by infrared, visible light and ultraviolet sensors, providing data support for subsequent evaluation of fire extinguishing effect and optimization of sprinkler strategy.

[0070] Step 7: The intelligent decision-making module, based on real-time fire source monitoring data under feedback monitoring status, dynamically evaluates and adaptively optimizes the effectiveness of fire extinguishing strategies. The specific process is as follows: During sprinkler operation, the intelligent decision-making module continuously collects real-time fire data of the fire source area using an infrared thermal imager, a visible light camera, and an ultraviolet flame detector. Within a preset time window (e.g., 10 seconds), it analyzes the temperature change trend of the flame root area. Simultaneously, based on the visible light flame color feature map and ultraviolet flame feature map continuously generated in step five (e.g., step 502), it analyzes the historical change trend of the signal intensity they represent. If the flame root temperature drops by more than a set threshold (e.g., 50%), and the visible light and ultraviolet signals weaken synchronously, the current strategy is deemed effective, and the intelligent decision-making module maintains the existing sprinkler control commands unchanged.

[0071] If, within a preset time window, the temperature drop in the flame root area does not exceed a threshold (e.g., 50%) or shows a continuous upward trend, or if the visible light flame color characteristic signal or ultraviolet flame detection signal continuously strengthens, it is determined that the fire is out of control or there is a large leak. The intelligent decision-making module immediately upgrades the fire extinguishing strategy according to the fire extinguishing rule base and generates an enhanced sprinkler control command. The enhanced sprinkler control command is used to control the opening of the solenoid valve of the corresponding protected sub-zone, so that the fixed sprinklers and adjustable sprinklers in the protected sub-zone are put into operation simultaneously. The adjustable sprinklers continuously spray the main fire source area in a directional manner, while the fixed sprinklers spray the entire protected sub-zone in a comprehensive manner, thereby forming a three-dimensional fire extinguishing situation that takes into account both local precision strikes and overall area coverage to suppress the spread of the fire. As an example, the enhanced sprinkler control command includes at least: increasing the number of adjustable sprinklers participating in the spray, opening the fixed sprinklers, adjusting the spray mode, and increasing the spray flow parameters. Commands such as adding the number of the enabled adjustable and fixed nozzles, switching the spray mode to a higher intensity "focused water jet" or expanding the "scan" range, and increasing the flow rate are all possible.

[0072] When the data collected by the infrared thermal imager, visible light camera, and ultraviolet flame detector all return to the normal operating threshold range, and no abnormal changes are detected within a stable period of time (e.g., 30 seconds), the intelligent decision module determines that the fire has been extinguished and generates a sprinkler termination control command. After receiving the sprinkler termination control command, the drive controller controls the adjustable sprinkler head to perform a return operation. At the same time, after receiving the sprinkler termination control command, the solenoid valve closes the valve, and the system immediately returns to the automatic monitoring state.

Claims

1. A smart directional spraying method for FPSO based on multi-sensor fusion, characterized in that... Includes the following steps: Step 1: Construct a protection sub-zone and integrate a multi-source fusion monitoring terminal within the FPSO upper module. Arrange the monitoring terminal and nozzles within the FPSO upper module and network them using an industrial wireless network. Install a solenoid valve and an electrically controlled flow regulating valve on the pipeline branch of each protection sub-zone. The monitoring terminal includes a fire detection sensor and an attitude sensor. The fire detection sensor includes an infrared thermal imager, an ultraviolet flame detector, and a visible light camera. Within each zone, a hybrid nozzle layout is adopted, including adjustable nozzles and fixed nozzles. The adjustable nozzles are connected to a rotary drive device. The drive controller of the rotary drive device, the solenoid valve of the protection sub-zone, and the electrically controlled flow regulating valve are collectively referred to as the execution unit. Step 2: Establish a module coordinate system and build a model of the FPSO hull in the edge computing server; Step 3: Perform spatial pose and detection parameter calibration on the three types of fire detection sensors: infrared thermal imager, visible light camera, and ultraviolet flame detector. The calibration yields the intrinsic and extrinsic parameter matrices of the infrared thermal imager and visible light camera, as well as the effective response area parameters of the ultraviolet detector. After calibration, the infrared images acquired by the infrared thermal imager and the visible light images acquired by the visible light imager are unified to the module coordinate system. At the same time, the effective response area of ​​the ultraviolet detector obtained from the calibration is unified to the module coordinate system through coordinate transformation to form spatial registration data, which is then stored in the edge computing server. Step 4: Based on the spatial registration data from Step 3, map the existing multimodal fire data to the module coordinate system. Use the mapped multimodal fire data to construct and train a convolutional neural network fire identification model that fuses multi-source information. The convolutional neural network fire identification model outputs structured results: three-dimensional spatial coordinates of the fire source, fire type, and fire spread direction vector. Step 5: After the system enters the running state, the edge computing server uses the ship attitude data collected by the attitude sensor to perform image stabilization processing on the infrared image data collected by the infrared thermal imager and the visible light image data collected by the visible light camera. Using the calibration parameters obtained in step three, input the stabilized infrared image and visible light image into the module coordinate system; The edge computing server extracts fire-related features from the stabilized infrared image, visible light image, and spatially registered ultraviolet flame detection signal, and constructs a multi-channel fusion feature tensor. A four-channel fusion feature tensor is then constructed and input into the trained convolutional neural network fire recognition model for real-time processing to obtain the fire situation recognition result. Finally, the intelligent decision-making module deployed on the edge computing server generates the corresponding fire extinguishing strategy based on the fire situation recognition result. Step Six: Based on the fire extinguishing strategy in Step Five, the edge computing server calculates the number of adjustable sprinkler heads to be activated to obtain the corresponding sprinkler control command. The sprinkler control command includes at least: fire type, activated adjustable sprinkler head number and fixed sprinkler head number, sprinkler mode, three-dimensional target coordinates of the fire source in the module coordinate system, and sprinkler flow rate level. Then, it is sent to the corresponding execution unit through the industrial wireless network in a time sequence. The execution unit performs the sprinkler fire extinguishing operation according to the fire extinguishing strategy and executes the sprinkler control command. After the execution unit starts execution, the intelligent decision module enters the feedback monitoring state and continuously receives real-time data of the fire source area collected by the infrared thermal imager, ultraviolet flame detector and visible light camera. Step 7: The intelligent decision-making module conducts dynamic evaluation and adaptive optimization control of the fire extinguishing strategy execution effect based on real-time fire source monitoring data under feedback monitoring status.

2. The FPSO intelligent directional spraying method based on multi-sensor fusion according to claim 1, characterized in that: Step one includes the following steps: Step 101: Fire compartmentation and sprinkler arrangement are carried out in the protected space to form a fire suppression capability that combines basic coverage with precise fire suppression; the details are as follows: The protection area of ​​the upper module of the FPSO is divided into multiple independent protection sub-zones and a fire water supply network is laid out. The fire water supply network is arranged in the structural interlayer of the ceiling of the module. A solenoid valve and an electrically controlled flow regulating valve are installed on the pipe branch of each protection sub-zone. Adjustable spray head: One adjustable spray head is arranged at the center of each protected sub-zone or directly above key equipment; the rotary drive device is a direct-drive dual motor, which includes a horizontal rotary stepper motor (4-1) mounted on the ceiling, with the output shaft of the horizontal rotary stepper motor set vertically downward; the body of the vertical rotary stepper motor (4-2) is fixedly connected to the output shaft of the horizontal rotary stepper motor, with the output shaft of the vertical rotary stepper motor set in the horizontal direction; the adjustable spray head (4-3) is fixedly connected to the output shaft of the vertical rotary stepper motor; Around the adjustable nozzle, multiple fixed nozzles are evenly arranged in a grid pattern, and the fixed nozzles are installed on the ceiling. Step 102: Distribute and install monitoring terminals to obtain real-time fire and ship attitude information; The fire detection sensor is installed inside the upper module of the FPSO on a rigid bracket 3 to 5 meters above the main deck. The installation position ensures that the combined field of view of the infrared thermal imager, ultraviolet flame detector and visible light camera can completely cover all the protective equipment areas downwards. The attitude sensor is used to measure the ship's roll angle in real time. and pitch angle The attitude sensor is rigidly mounted on the main steel structure of the upper module of the FPSO; Step 103: Connect all monitoring terminals and execution units to the edge computing server located in the secure area via the industrial wireless network to form a system control loop.

3. The FPSO intelligent directional spraying method based on multi-sensor fusion according to claim 2, characterized in that: The low, medium, and high speeds of the electronically controlled flow regulating valve correspond to 20%-40%, 60%-80%, and 95%-100% of the rated flow of a single adjustable nozzle, respectively.

4. The FPSO intelligent directional spraying method based on multi-sensor fusion according to claim 3, characterized in that: The adjustable and fixed nozzles are explosion-proof nozzles, and the edge computing server is an explosion-proof edge computing server.

5. The FPSO intelligent directional spraying method based on multi-sensor fusion according to claim 2, characterized in that: Step three specifically includes the following steps: Step 301, Visual sensor calibration data acquisition: The infrared thermal imager and the visible light camera are calibrated using the Zhang Zhengyou planar calibration method, which specifically includes: taking multiple images of the standard calibration board in different postures using the infrared thermal imager and the visible light imaging head, and obtaining image data for calibration. Step 302: The calibration parameters of the infrared thermal imager and the visible light camera are calculated by the calibration algorithm and then input into the edge computing server. The calibration parameters of the infrared thermal imager and the visible light camera include the intrinsic parameter matrix K of the infrared thermal imager and the visible light camera, which includes focal length parameters, principal point coordinates and distortion parameters; and the extrinsic parameter matrices of the infrared thermal imager and the visible light camera relative to the module coordinate system, which include rotation matrix R and translation vector T, used to describe the spatial position and attitude of the infrared thermal imager and the visible light camera in the module coordinate system. Step 303: Calibrate the effective response region of the ultraviolet detector in the module coordinate system and store it in the edge computing server; Step 304, Spatial registration of fire detection sensors: Based on the intrinsic and extrinsic parameter matrices of the infrared thermal imager and the visible light camera obtained from calibration, the infrared images acquired by the infrared thermal imager and the visible light images acquired by the visible light imager are unified to the module coordinate system through coordinate transformation and image reprojection technology. At the same time, the effective response area of ​​the ultraviolet detector obtained from calibration in step 303 is unified to the module coordinate system through coordinate transformation.

6. The FPSO intelligent directional spraying method based on multi-sensor fusion according to claim 5, characterized in that: Step five specifically includes the following steps: Step 501, Monitoring terminal data synchronous acquisition and image stabilization preprocessing of visual sensor output images: During system operation, data from the infrared thermal imager, visible light camera, ultraviolet flame detector, and attitude sensor are synchronously acquired at preset time intervals. After aligning the data from the infrared thermal imager and visible light camera with timestamps, the data is transmitted to the edge computing server, as detailed below: First, the ship's attitude rotation matrix is ​​calculated: the ship's roll angle is obtained in real time using attitude sensors. and pitch angle We construct the basic rotation matrices of the FPSO hull around the X-axis and the Y-axis respectively; then we calculate the hull attitude rotation matrix at the k-th sampling time based on the above basic rotation matrices. Subsequently, the hull attitude rotation matrix was used. Digital image stabilization is performed on infrared and visible light images. For any pixel in the infrared and visible light images, combined with the calibration intrinsic and extrinsic parameter matrices obtained in step three, the pixel is mapped to a stable module coordinate system through coordinate transformation. Finally, after completing the above coordinate transformation, the spatial point corresponding to any pixel in the infrared image and visible light image in the module coordinate system is reprojected onto the virtual horizontal image plane to obtain the stabilized infrared image and visible light image, which are stored in the edge computing server, thereby eliminating the pixel offset caused by the hull sway. Step 502: Based on the image stabilization process, the edge computing server extracts fire-related features from the stabilized infrared image, visible light image, and spatially registered ultraviolet flame detection signal, and constructs a multi-channel fusion feature tensor as input to the subsequent fire identification model. The specific steps are as follows: Infrared temperature feature map construction, including: Based on the stabilized infrared image acquired by the infrared thermal imager, pixel-by-pixel temperature information is extracted and normalized to obtain an infrared temperature feature map. This infrared temperature feature map is used to characterize the high-temperature distribution characteristics of suspected fire source areas; Construction of Visible Light Flame Color Feature Map: The stabilized visible light image acquired by the visible light camera is converted from the RGB color space to the HSV color space, and a flame color likelihood map is constructed based on typical flame color features. Visible light flame color feature map and motion feature map are generated respectively. The visible light flame color feature map is used to reflect the distribution of regions with flame color features in the visible light image. Motion feature map construction: Motion feature maps are extracted from stabilized visible light images acquired by a visible light camera using the inter-frame difference method. This motion feature map is used to describe the dynamic characteristics of a flame as it changes over time; Ultraviolet Flame Feature Map Construction: During real-time system operation, a binary ultraviolet flame feature map UVmap(i,j) is generated based on the alarm status of the ultraviolet detector. This binary ultraviolet flame feature map visually represents the ultraviolet detector's judgment result on the presence and spatial location of the flame in an image-like form. The generation rules are as follows: When the ultraviolet detector does not trigger an alarm, all pixel values ​​in UVmap(i,j) are set to 0. When the ultraviolet detector alarms, all pixels (i, j) within its effective response area in UVmap(i,j) are assigned a value of 1, and all pixels in the remaining area are assigned a value of 0. Multi-channel fusion feature tensor construction: The infrared temperature feature map, visible flame color feature map, motion feature map, and ultraviolet flame feature map obtained above are spatially aligned and then stacked according to channel dimensions to form a four-channel fusion input feature tensor Xin. ; Step 503, Fire identification and fire situation analysis based on a convolutional neural network fire identification model: The multi-channel fusion feature tensor Xin obtained in step 502 is used The data is input into a pre-trained and loaded convolutional neural network fire identification model to perform fire identification and situation analysis; the convolutional neural network fire identification model outputs a structured fire situation identification result, which includes at least: Fire source location information, fire attribute information, and fire dynamic information. Step 504: Generation of fire extinguishing strategy: First, the intelligent decision-making module deployed on the edge computing server receives the fire situation identification result output in step 503. The intelligent decision-making module integrates a fire extinguishing rule base. Based on the fire situation identification result, the intelligent decision-making module generates a fire extinguishing strategy for the current fire situation by performing rule matching and reasoning on the fire extinguishing rule base. The fire extinguishing rule base pre-stores fire extinguishing strategy rules corresponding to different fire source location information, fire attribute information, and fire dynamic information. The fire extinguishing rule base includes at least the following strategies: For pool fire: Activate the 2-3 adjustable nozzles closest to the fire source and adopt a medium-speed scanning spray mode—control the adjustable nozzles to perform horizontal reciprocating scanning motion at a constant speed under the calculated vertical rotation angle, and at the same time set the electronically controlled flow regulating valve of the protection sub-zone to "medium" to cover and suppress the burning liquid surface area; For jet fire: Activate 1-2 adjustable nozzles upstream of the fire source and use the focused water jet mode—control the nozzles to maintain a fixed angle and accurately aim at the root of the flame. At the same time, set the electronically controlled flow regulating valve of the protected sub-zone to "high" for cooling and shut-off. For electrical fires: The system will trigger the power supply to the equipment in the area. After confirming that the power outage conditions are met, the adjustable sprinkler head above the fire source will be activated and a fine water mist mode will be adopted. The sprinkler head will be controlled to slowly and slightly oscillate to cover the protected equipment. At the same time, the electronically controlled flow regulating valve of the protected sub-zone will be set to "low" to cool down and extinguish the fire. The medium-speed scanning spray mode, focused water column mode, and fine water mist mode are three spray modes of the adjustable nozzle.

7. The FPSO intelligent directional spraying method based on multi-sensor fusion according to claim 6, characterized in that: Step six specifically includes the following steps: Step 601, Core calculation of sprinkler control commands: Spatial calculation of fire source location and nozzle installation coordinates: calling all adjustable nozzle installation coordinates stored in the FPSO hull model ( , , ), P =1, 2, ..., K, where K is the total number of adjustable nozzles. The spatial distance between each adjustable nozzle and the three-dimensional spatial coordinates of the fire source is calculated based on the Euclidean distance formula. The N adjustable nozzles that are closest to the fire source are selected as the target adjustable nozzles to be activated. Matching calculation of fire type and sprinkler flow rate: Based on the preset flow rate level of the fire extinguishing strategy, and combined with the infrared temperature feature map, the average temperature of the flame root region is extracted as the judgment criterion, denoted as... ;like >400℃, increase the strategy level by 1 level; if 200℃ ≤ ≤400℃, execute according to the strategy setting; if If the temperature is below 200℃, the original strategy level will be uniformly lowered to the lowest level. Optimization calculation of the number of adjustable nozzles to be activated: based on the fire spread direction vector in the fire suppression strategy. ( , , ),in , , These represent the components of the fire spread direction vector along the X, Y, and Z axes of the module coordinate system, combined with the average temperature at the flame root. Select the target adjustable nozzles to be turned on; if >400℃, a maximum of 3 target adjustable nozzles can be activated; if 200℃≤ At ≤400℃, a maximum of two target-adjustable nozzles can be activated; <200℃, only one target adjustable nozzle is activated, and the adjustable nozzle installation coordinates in the FPSO hull model from step two are called ( , , First, select adjustable nozzles that are ≤8 meters away from the fire source in a straight line, and then further select them according to direction: When =0, select the adjustable nozzle that is closest to the fire source in the X-axis direction; When =0, select the nozzle that is closest to the fire source in the Y-axis direction; , When none of them are equal to 0, select the adjustable nozzles that are closest to the main direction of fire spread and to both sides. If the fire source is ≤2 meters from the boundary of the protected sub-zone, open one additional adjustable nozzle in the adjacent sub-zone. The above calculations generate standardized sprinkler control commands; Step 602, Instruction Issuance and Spraying Operation Execution: The edge computing server sends sprinkler control commands to the drive controllers and electronically controlled flow regulating valves of the protection sub-zones corresponding to the adjustable sprinkler heads. After receiving the instruction from the edge computing server, the drive controller calculates the required rotation angle for the adjustable nozzle to align with the fire source based on inverse kinematics calculations and the installation coordinates of the nozzle in the module coordinate system. Then, according to the spray mode in the instruction, it calls the pre-stored motion control algorithm corresponding to that spray mode to complete the nozzle attitude pre-adjustment: in the focused water column mode, the drive horizontal and vertical stepper motors rotate horizontally by an angle... Vertical turning angle Precise rotation ensures the nozzle axis stably points towards the target point; in medium-speed scanning mode, the pre-stored trajectory planning algorithm in the drive controller is invoked to fix the vertical angle of the adjustable nozzle. The horizontal angle of the adjustable nozzle can be adjusted to... Using the baseline value, within the preset scan amplitude Δ The system generates a continuous reciprocating scanning trajectory and uses an interpolation algorithm to achieve uniform nozzle oscillation. In fine water mist mode, it calls a pre-stored micro-motion control algorithm in the drive controller and uses the horizontal rotation angle obtained from the aforementioned inverse kinematics calculation. Vertical turning angle Using the angle as a reference, a tiny periodic oscillation trajectory is generated, causing the nozzle to... and Small, slow movements nearby create a fine mist covering the area; The electronically controlled flow regulating valve receives the flow level instruction from the edge computing server and adjusts the valve opening to near the target level. After the nozzle posture and flow rate are adjusted to the correct positions, the edge computing server sends the spray control command to the solenoid valve of the protected sub-area, and the solenoid valve performs the corresponding on / off control. After the execution unit completes the above actions, the intelligent decision-making module enters the feedback monitoring state, that is, it continuously receives real-time data of the fire source area collected by infrared, visible light and ultraviolet sensors.

8. The FPSO intelligent directional spraying method based on multi-sensor fusion according to claim 7, characterized in that: Step five specifically includes the following steps: During the sprinkler operation, the intelligent decision-making module continuously collects real-time fire data of the fire source area through an infrared thermal imager, a visible light camera, and an ultraviolet flame detector. Within a preset time window, it analyzes the temperature change trend of the flame root area. At the same time, based on the visible light flame color feature map and ultraviolet flame feature map continuously generated in step five, it analyzes the historical change trend of the signal intensity they represent. If the temperature at the flame root drops beyond a set threshold and the visible light and ultraviolet signals weaken synchronously, the current strategy is deemed effective, and the intelligent decision-making module maintains the existing sprinkler control command unchanged. If, within a preset time window, the temperature drop in the flame root area does not exceed the threshold or shows a continuous upward trend, or if the visible light flame color characteristic signal or the ultraviolet flame detection signal continuously strengthens, it is determined that the fire is out of control or there is a large leak. The intelligent decision-making module immediately upgrades the fire extinguishing strategy according to the fire extinguishing rule base and generates an enhanced sprinkler control command. The enhanced sprinkler control command is used to control the opening of the solenoid valve of the corresponding protection sub-zone, so that the fixed nozzles and adjustable nozzles in the protection sub-zone can work at the same time. The adjustable nozzles continuously spray the main fire source area in a directional manner, while the fixed nozzles spray the entire protection sub-zone in a covering manner. When the data collected by the infrared thermal imager, visible light camera, and ultraviolet flame detector all return to the normal operating threshold range and no abnormal changes are detected within a stable period of time, the intelligent decision module determines that the fire has been extinguished and generates a sprinkler termination control command. After receiving the sprinkler termination control command, the drive controller controls the adjustable sprinkler head to perform a return operation. At the same time, after receiving the sprinkler termination control command, the solenoid valve closes the valve, and the system immediately returns to the automatic monitoring state.

Images