Intelligent safety control system for cargo hold of HBI carrier and method thereof

Abstract

The application discloses an HBI transport ship cargo hold intelligent safety prevention and control system, which comprises a multi-parameter sensor system, a multi-parameter fusion pre-judging module and a hierarchical linkage emergency module. Through a three-dimensional fusion model of a temperature gradient, hydrogen speed increase and ship roll angle which is constructed based on a BP neural network, and in cooperation with an early warning time extrapolation algorithm of a coupling correction coefficient alpha, accurate pre-judgment of HBI self-heating risks is realized. Compared with a traditional threshold alarm, the early warning accuracy is improved by 80%, the false alarm rate is less than 5%, the missed alarm rate is less than 1%, and the early warning advance time is greater than or equal to 30 minutes.

Description

Technical Field

[0001] This invention relates to HBI (Hardware-Based Insurance) transportation security technology, and more particularly to an HBI intelligent security control system and method for cargo holds on transport ships. Background Technology

[0002] Hot Briquetted Iron (HBI), as an important industrial raw material, plays a crucial role in modern steel production. With the development of the global steel industry, the production and trade volume of HBI have increased year by year, and its demand for maritime transport has also grown. However, HBI possesses many unique properties and potential risks during transportation, posing a serious challenge to the safety of ship transport.

[0003] HBI falls under the category of "Direct Reduced Iron (A)" in the International Marine Solid Bulk Code (IMSBC Code): a gray metallic substance produced by a densification process and molded into brick shapes. According to the IMSBC Code's description of the hazards of "Direct Reduced Iron (A)," its temperature is expected to temporarily rise by approximately 30°C due to self-heating after loading and unloading in bulk. This substance may slowly release hydrogen upon contact with water (especially saltwater).

[0004] After being loaded with bulk hydrogen biogas (HBI), the HBI may generate heat within a short period in the cargo hold, raising its temperature to 50°C. When HBI comes into contact with water, especially saline water or seawater, the oxidation-reduction reaction is accelerated. The large amount of heat released further intensifies the oxidation reaction, creating a chain reaction. Under sufficient oxygen conditions, the temperature can reach 1500°C. In a confined space (such as a ship's hold), the released hydrogen is a flammable gas. When it mixes with air, and the concentration (by volume) exceeds 4%, an explosive mixture is formed. This can ultimately lead to safety problems such as asphyxiation, spontaneous combustion, and explosion.

[0005] Currently, there are no specially designed transport vessels for HBI (Hydrogen Inhalation Biohazard) and conventional bulk carriers are still used for transportation. Furthermore, existing ship cargo hold monitoring systems are mostly based on single alarm modes using real-time thresholds, only able to respond immediately to hazards that have already occurred—a "passive triggering" type of early warning. These systems cannot predict the accumulation trend of hydrogen concentration under unventilated conditions several hours in advance, leading to a lack of quantitative decision-making basis for crew members when ventilation vents need to be closed for extended periods in severe weather. They also cannot intervene early before the cargo spontaneous combustion threshold, often only issuing warnings after the danger has already occurred. In addition, existing emergency responses mostly rely on manual judgment and operation, with response times typically exceeding 10 minutes, far from meeting the rapid evolution requirements of HBI spontaneous combustion and hydrogen explosion. Summary of the Invention

[0006] To address the aforementioned problems in the prior art, this invention provides an HBI intelligent safety control system and method for cargo holds of transport ships, which can perform multi-parameter fusion prediction and can respond promptly through hierarchical linkage.

[0007] On the one hand, an HBI (Intelligent Safety and Security) system for cargo holds of transport ships includes: A multi-parameter sensor system collects real-time data on temperature gradient, hydrogen concentration, oxygen concentration, and ship attitude within the cargo hold. The multi-parameter fusion prediction module has a built-in real-time fusion prediction unit based on BP neural network. The real-time fusion prediction unit takes the temperature difference rate between the upper and lower layers of the cargo hold, the hydrogen concentration change rate, and the ship's roll angle as input parameters. After feature extraction by the double hidden layer, the Softmax layer outputs the probability distribution of five risk levels and calculates the warning time based on the extrapolation model of the coupling correction coefficient to ensure that the warning is given at least 30 minutes in advance. The tiered emergency response module receives the output signal from the multi-parameter fusion prediction module and executes tiered handling instructions according to preset logic.

[0008] The multi-parameter sensor system includes temperature sensors respectively installed at the bottom and top of the cargo hold, hydrogen concentration sensors and oxygen sensors installed at the top of the cargo hold, and ship attitude sensors installed on the rigid structure in the middle of the cargo hold.

[0009] The multi-parameter fusion prediction module uses a four-layer BP neural network with three nodes in the input layer. It receives the three-dimensional input feature vector X=[ΔT_norm,ΔH2_norm,θ_norm] after Z-score normalization. Where ΔT_norm is the normalized rate of temperature difference between the upper and lower layers, ΔH2_norm is the normalized rate of hydrogen concentration change, and θ_norm is the normalized ship roll angle; the first hidden layer has 12 nodes, uses the ReLU activation function, and outputs h. (1) =ReLU(W (1) ·X+b (1) ), where W (1) The weight matrix is ​​12×3, b (1) The bias vector is 12×1; the second hidden layer has 8 nodes, uses the ReLU activation function, and outputs h. (2) =ReLU(W (2) ·h (1) +b (2) ), where W (2) The weight matrix is ​​8×12, b (2) The bias vector is 8×1; the output layer has 5 nodes, uses the Softmax activation function, and the output Risk_Level = Softmax(W(3) ·h (2) +b (3) ), where W (3) The weight matrix is ​​5×8, b (3) The bias vector is 5×1. Each element of the output vector corresponds to the predicted probability of the five risk levels. The sum of all elements is 1. The category with the highest probability is taken as the output risk level.

[0010] The formula for calculating the warning time Tpredict is:

[0011] In the formula, T predict Threshold is the warning time. critical The critical threshold is set to either 1%vol for hydrogen concentration or 100℃ for temperature. Current_Value is the current measured value, dX / dt is the parameter change rate, and α(ΔH2, θ) is the coupling correction coefficient.

[0012] The coupling correction coefficient α is determined according to the following rules: when ΔH2 ≥ 1.0%vol / h, α = 0.8; when θ ≥ 15°, α = 0.7; when ΔH2 ≥ 1.0%vol / h and θ ≥ 15°, α = 0.6; and under normal operating conditions, α = 1.0.

[0013] The input-output logic decision matrix of the multi-parameter fusion prediction module is as follows: when ΔT < 3℃ / h, ΔH2 < 0.1%vol / h, and θ < 10°, output Level 0, system standby monitoring; when ΔT ≥ 3 and < 5℃ / h, ΔH2 < 0.5%vol / h, and θ < 10°, output Level 1, warning time ≥ 30min; when ΔT ≥ 5℃ / h, ΔH2 is between 0.5-1.0%vol / h, and θ < 10°, output Level 2, warning time ≥ 30min; when ΔT ≥ 5℃ / h or temperature ≥ 100℃, ΔH2 is between 1.0-4.0%vol / h, and θ ≥ 10°, output Level 3, warning time ≥ 30min; when temperature ≥ 150℃, ΔH2 ≥ 4.0%vol / h, and θ ≥ 15°, output Level 4.

[0014] The tiered emergency response module includes four levels of linkage units. Its triggering condition is determined by a logic gate composed of the risk level output by the multi-parameter fusion prediction module and the sensor's measured threshold. Specifically, it includes: Level 1 early warning unit: When the risk level is Level 1 and ΔT≥3℃ / h and H2≤0.5%vol, increase the sampling frequency of the multi-parameter sensor system to 1 time / 10 seconds; Level 2 warning unit: When the risk level is Level 2 and ΔT≥5℃ / h and H2 is between 0.5% and 1.0%vol, the forced ventilation at the top of the cargo hold will be automatically activated, with a ventilation rate of 3 air changes per hour; Level 3 alarm unit: When the risk level is Level 3 and H2 is between 1.0% and 4.0% vol or the temperature is ≥100℃, the nitrogen inerting system is activated, injecting nitrogen into the cargo hold at a standard flow rate of 50 m³ / h to reduce the oxygen volume content in the hold to below 5% within 30 seconds to suppress combustion. Afterward, nitrogen is intermittently replenished through closed-loop feedback from the oxygen sensor to maintain the oxygen supply. During this time, the explosion-proof pressure relief vents in the cargo hold remain closed. Level 4 Emergency Unit: When the risk level is Level 4 and H2 ≥ 4.0% vol or temperature ≥ 150℃, the following actions shall be performed simultaneously: increase the nitrogen injection rate to an enhanced flow rate of 150 m³ / h, reduce the oxygen volume content in the cabin to below 2% within 60 seconds to suppress the explosion, and perform a forced inerting cycle throughout the cabin; open all explosion-proof pressure relief vents in the cargo hold; cut off all non-explosion-proof power supplies around the cargo hold; and lock the cargo hold hatch covers.

[0015] The response sequence of the hierarchical linkage emergency module is as follows: signal transmission delay <100ms, controller processing cycle <50ms, actuator mechanical response <3s, and total response time from signal acquisition to linkage execution <5 seconds.

[0016] It also includes a redundant communication module and a data storage unit. The redundant communication module includes a CAN bus wired communication unit and a 5G wireless communication unit, with an end-to-end data latency of <100ms. The data storage unit is used to store no less than 6 months of continuous monitoring data.

[0017] On the other hand, a prevention and control method for an HBI (Intelligent Safety and Security) cargo hold system for transport ships includes the following steps: A. Real-time data collection of temperature gradient, hydrogen concentration, oxygen concentration, and ship attitude data within the cargo hold using a multi-parameter sensor system; B. Through a multi-parameter fusion prediction module with a built-in real-time fusion prediction unit based on a BP neural network, the temperature difference rate between the upper and lower layers of the cargo hold, the rate of change of hydrogen concentration, and the ship's roll angle are used as input parameters. After feature extraction by the double hidden layer, the Softmax layer outputs the probability distribution of five risk levels, and the warning time is calculated based on the extrapolation model of the coupling correction coefficient to ensure that the warning is given at least 30 minutes in advance. C. The hierarchical linkage emergency module receives the output signal of the multi-parameter fusion prediction module and executes hierarchical handling instructions according to preset logic.

[0018] The HBI intelligent safety control system and method for cargo holds of transport ships, as described in this invention, has the following advantages: 1. A three-dimensional fusion model of temperature gradient, hydrogen gas acceleration, and ship roll angle, constructed based on a BP neural network (12×8 structure with dual hidden layers and Softmax five-class output), combined with an early warning time extrapolation algorithm with coupling correction coefficient α, was developed to accurately predict HBI self-heating risk. The early warning accuracy rate is 80% higher than that of traditional threshold alarms, with a false alarm rate of <5%, a missed alarm rate of <1%, and an early warning lead time of ≥30 minutes.

[0019] 2. A four-level gradient response mechanism strictly corresponding to the hazardous nature of HBI cargo has been established, with trigger control through a dual-input logic gate of "model prediction + real-time threshold". At level three alarm, the oxygen content is reduced to below 5% within 30 seconds to suppress combustion; at level four emergency, the oxygen content is reduced to below 2% within 60 seconds, combined with full-cabin inerting circulation and explosion-proof depressurization. The entire response time is less than 5 seconds, meeting the requirements of ship safety regulations. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the prevention and control system of the present invention; Figure 2 This is a schematic diagram of the BP neural network model of the present invention. Detailed Implementation

[0021] The following is a further description of an HBI (Intelligent Safety and Security Control) system for cargo holds of transport ships according to the present invention. For example... Figure 1 As shown, the HBI intelligent safety control system for cargo holds of transport ships of the present invention includes a multi-parameter sensor system, a multi-parameter fusion prediction module, and a hierarchical linkage emergency module. The multi-parameter fusion prediction module is electrically connected to the multi-parameter sensor system and the hierarchical linkage emergency module, respectively. The multi-parameter sensor system is used to collect real-time data on temperature gradient, hydrogen concentration, oxygen concentration, and ship attitude within the cargo hold. The multi-parameter sensor system includes temperature sensors deployed at the bottom and top of the cargo hold, hydrogen concentration and oxygen sensors deployed at the top of the cargo hold, and ship attitude sensors (such as IMUs) deployed on a rigid structure in the middle of the cargo hold. A specific arrangement can be shown in Table 1. Table 1

[0022] Temperature sensors are deployed at least one set per 50m² in the heat accumulation zone at the bottom of the cargo hold and at the corresponding top, to monitor the temperature difference between the upper and lower layers caused by the self-heating of the HBI. Hydrogen concentration sensors are deployed at least one set per 80m² in the hydrogen top accumulation zone at the top of the cargo hold to monitor the hydrogen concentration released by the HBI oxidation reaction. Oxygen sensors are deployed at least one per 160m² at the top of the cargo hold for oxygen content feedback control of the inerting system. One ship attitude sensor is installed on the rigid structure in the middle of each cargo hold to collect roll and pitch angle data. Additionally, one set of humidity sensors for monitoring the corrosive environment can be deployed at the bottom and top of each cargo hold.

[0023] The multi-parameter fusion prediction module is the core computing and decision-making unit of the prevention and control system. It incorporates a real-time fusion prediction unit for HBI self-heating risk based on a BP neural network. This real-time fusion prediction unit uses the temperature difference between the upper and lower cargo holds, the rate of change in hydrogen concentration, and the ship's roll angle as input parameters, obtained from the multi-parameter sensor system output. After feature extraction via a double hidden layer, the Softmax layer outputs the probability distribution of five risk levels. Based on an extrapolation model with coupling correction coefficients, it calculates the warning time to ensure that the warning is issued at least thirty minutes in advance.

[0024] Its principle is as follows: Figure 2 As shown: The model input layer receives three dynamic feature vectors after Z-score normalization: the rate of change of temperature difference ΔT (unit: °C / h) representing the degree of thermal stratification in the vertical direction of the cargo hold; the rate of change of hydrogen concentration ΔH2 (unit: %vol / h) representing the chemical activity of HBI; and the ship's roll angle θ (unit: degrees) representing the physical displacement of the cargo stack and the gas stratification disturbance state. The input data is preprocessed using moving average filtering (5-minute window) and outlier removal (3σ criterion).

[0025] See Table 2 for details: Table 2

[0026] Model structure: The BP neural network adopts a four-layer structure, including an input layer, a first hidden layer, a second hidden layer, and an output layer. See Table 3 for details. Table 3

[0027] The input layer has 3 nodes, corresponding to the 3D input feature vector X=[ΔT_norm, ΔH2_norm, θ_norm] after Z-score normalization. ΔT_norm is the normalized rate of change of temperature gradient, ΔH2_norm is the normalized rate of change of hydrogen concentration, and θ_norm is the normalized ship roll angle. The first hidden layer has 12 nodes and uses the ReLU activation function to learn the nonlinear coupling relationship between parameters. The second hidden layer has 8 nodes and uses the ReLU activation function to reduce dimensionality and extract key risk features. The output layer has 5 nodes and uses the Softmax activation function to output the probability distribution vectors of the five risk levels.

[0028] Its mapping function is:

[0029] The forward propagation calculation process of the BP neural network is as follows: Step 1: Calculate the output h of the first hidden layer. (1) = ReLU(W (1) ·X+b (1) ), where W (1) The weight matrix is ​​12×3, b (1) The bias vector is 12×1; the second step is to calculate the output h of the second hidden layer. (2) =ReLU(W (2) ·h (1) + b (2) ), where W (2) The weight matrix is ​​8×12, b (2) The bias vector is 8×1; Step 3: Calculate the output layer risk level probability distribution: Risk_Level = Softmax(W (3) ·h (2) + b (3) ), where W (3) The weight matrix is ​​5×8, b (3) The bias vector is 5×1; the output vector Risk_Level is a 5×1 vector, with each element corresponding to the predicted probability of Level 0 (normal), Level 1 (attention), Level 2 (warning), Level 3 (alarm), and Level 4 (emergency). The sum of all elements is 1, and the system takes the category with the highest probability as the current output risk level.

[0030] Model output: See Table 4 for details. Table 4

[0031] Risk Level: Five-category output, namely Level 0 (Normal), Level 1 (Attention), Level 2 (Warning), Level 3 (Alarm), and Level 4 (Emergency).

[0032] Warning time Tpredict: Calculated based on an extrapolation model, representing the estimated remaining time from the current state to the critical state. The calculation formula is as follows:

[0033] In the formula: Tpredict is the warning time in minutes (min), and the system is configured to ensure that Tpredict ≥ 30 minutes; Threshold_ critical To trigger the critical threshold for the corresponding risk level, a hydrogen concentration critical value of 1%vol or a temperature critical value of 100℃ is used; Current_Value is the real-time parameter value collected by the sensor at the current moment; dX / dt is the rate of change of the key risk parameter at the current moment over time, in ℃ / h or %vol / h, after being processed by moving average filtering; α(ΔH2,θ) is the coupling correction coefficient, which is a dynamic correction factor based on the hydrogen concentration change rate ΔH2 (unit: %vol / h) and the ship roll angle θ (unit: degrees), used to quantify the superimposed accelerating effect of rapid hydrogen release and violent ship rolling on the risk evolution process.

[0034] The coupling correction coefficient α is determined according to the following rules: when ΔH2 ≥ 1.0% vol / h and θ < 15°, α = 0.8, indicating that rapid hydrogen release accelerates the risk process; when ΔH2 < 1.0% vol / h and θ ≥ 15°, α = 0.7, indicating that violent shaking disrupts gas stratification and accelerates the risk of mixed explosion; when ΔH2 ≥ 1.0% vol / h and θ ≥ 15°, α = 0.6, indicating that the two factors are superimposed and the risk process is significantly accelerated; when ΔH2 < 1.0% vol / h and θ < 15°, α = 1.0, indicating normal operating conditions with no additional acceleration effect.

[0035] The input-output logic decision matrix is ​​shown in Table 5: Table 5

[0036] Its core logical characteristic (non-linear coupling relationship) is: 1. Temperature is the dominant factor, with ΔT being the most sensitive indicator. A level 2 warning is triggered if the temperature is ≥5℃ / h. However, ΔH2 verification is required to rule out false positives (such as surface heating caused by solar radiation). 2. Hydrogen verification: ΔH2 ≥ 0.5% vol / h confirms that HBI has entered the oxidative autothermal stage; the hydrogen growth rate is exponentially related to the temperature difference (Arrhenius equation), indicating nonlinear amplification; 3. Roll modulation: θ≥10° disrupts gas stratification, causing hydrogen at the top to mix with air and lowering the lower explosive limit (LEL); it also affects heat dissipation, with the risk level weighting coefficient increasing by 1.3 for every 5° increase in θ. 4. Time prediction: Based on BP neural network time series prediction, the time to the critical threshold is extrapolated by the current parameter change rate to ensure an advance of ≥30 minutes.

[0037] Model Training and Validation: The training set includes over 1200 sets of historical HBI self-heating case data (derived from shipping accident reports and laboratory simulations) and over 5000 sets of multiphysics simulation data (CFD thermal-fluid coupling and chemical reaction kinetics simulation); the validation set includes over 800 sets of real-ship monitoring data for six consecutive months. Model performance indicators are: 80% improvement in early warning accuracy compared to traditional threshold alarms; false alarm rate <5% (suppressed by a hydrogen acceleration validation mechanism to prevent false temperature positives); missed alarm rate <1% (three-parameter cross-validation, degraded operation when a single parameter fails); early warning time ≥30 minutes. See Table 6 for details. Table 6

[0038] The tiered emergency response module is connected to the output of the multi-parameter fusion prediction module, receives the output signal from the multi-parameter fusion prediction module, and executes tiered response instructions according to preset logic. The tiered emergency response module includes four-level linkage units, whose triggering conditions are determined by a logic gate composed of the risk level output by the multi-parameter fusion prediction module and the sensor's measured threshold. The total response delay from signal acquisition to linkage execution is less than 5 seconds, with signal transmission delay (CAN bus + 5G dual redundancy) <100ms, PLC hard logic judgment processing cycle <50ms, and actuator (solenoid valve / fan / nitrogen valve) mechanical response <3s.

[0039] The triggering conditions of the linkage mechanism strictly match the physicochemical properties of HBI, and the judgment is made by a logic gate composed of the risk level output by the multi-parameter fusion prediction module and the sensor's measured threshold. Level 1 Early Warning Module: When the risk level is Level 1 (attention) and the real-time temperature difference ΔT ≥ 3℃ / h and the hydrogen concentration ≤ 0.5%vol, the sensor sampling frequency will be increased from 1 time / minute to 1 time / 10 seconds to perform high-frequency monitoring.

[0040] Level 2 warning module: When the risk level is Level 2 (warning) and ΔT≥5℃ / h and the hydrogen concentration is between 0.5%vol and 1.0%vol, the forced ventilation at the top of the cargo hold will be automatically activated, with a ventilation rate of 3 times / hour.

[0041] Level 3 Alarm Module: When the risk level is Level 3 (alarm) and the hydrogen concentration is between 1.0%vol and 4.0%vol, or the cargo stack temperature is ≥100℃, the nitrogen inerting system is activated. The standard nitrogen injection flow rate is 50 m³ / h, reducing the oxygen volume content in the cargo hold to below 5% within 30 seconds to suppress combustion; after reaching the target oxygen content, intermittent nitrogen replenishment is used to maintain it, with closed-loop feedback control via an oxygen sensor; the explosion-proof pressure relief vent remains closed during this time.

[0042] Level 4 Emergency Module: When the risk level is Level 4 (emergency) and the hydrogen concentration is ≥4.0% vol or the cargo stack temperature is ≥150℃, simultaneously execute all of the following actions: Increase the nitrogen injection rate to an enhanced flow rate of 150 m³ / h (3 times the standard flow rate), target oxygen content ≤2% to suppress explosion (far below the hydrogen explosion limit of 4% vol), achieve the target within 60 seconds and perform a full-capacity forced inerting cycle; open all explosion-proof pressure relief vents to prevent structural overpressure; cut off all non-explosion-proof power supplies around the cargo hold; lock the cargo hold hatch covers to prevent air backflow. See Table 7 for details: Table 7

[0043] The prevention and control system of this invention also includes a redundant communication module and a data storage unit. The redundant communication module includes a CAN bus wired communication unit and a 5G wireless communication unit, forming a dual-redundant data transmission link to resist electromagnetic interference during ship navigation, ensuring reliable transmission of sampling commands and monitoring data, with an end-to-end data latency of <100ms. The data storage unit stores at least six months of continuous monitoring data, meeting IMO data preservation requirements. The prevention and control system can also support shore-based remote data synchronization via maritime satellite or 5G network, facilitating data analysis and accident tracing after the voyage.

[0044] Example 1 On a 60,000-ton deadweight HBI transport ship, the following hardware deployment is implemented: Temperature sensor groups are installed at a density of 50 m² in the heat accumulation zone at the bottom of the cargo hold, 120 mm above the inner floor plate, to monitor the temperature difference between the upper and lower layers caused by HBI self-heating. Hydrogen concentration sensors are installed at a density of 80 m² in the hydrogen top accumulation zone at the top of the cargo hold, 400 mm below the deck, to monitor the hydrogen concentration released by the HBI oxidation reaction. Oxygen sensors are installed at a density of one per 160 m² at the top of the cargo hold for closed-loop feedback control of the inerting system. One ship attitude sensor (IMU) is installed on the rigid structure in the middle of each cargo hold to collect roll and pitch angle data. Sensor brackets are wall-mounted, and penetration through the outer plating is strictly prohibited (to prevent seawater leakage). Sensors installed at the bottom are 100-150 mm above the inner floor plate (to avoid cargo impact and ensure airflow). Sensors installed at the top are 300-500 mm below the deck (the height of the hydrogen accumulation layer). All cables passing through the cabin are protected by galvanized steel pipes with fire-resistant sealing filler (A-60 fire rating). Grounding resistance ≤4Ω to prevent static electricity buildup.

[0045] All sensors are connected to the host computer in the ship's central control room via a CAN bus (i.e., the hardware carrier of the multi-parameter fusion prediction module), and at the same time, they maintain data synchronization with the shore-based center via a 5G wireless module.

[0046] Training and Operation of the BP Neural Network Model: In the laboratory, self-thermal oxidation simulation experiments were conducted using HBI samples at different moisture contents and ambient temperatures. This was combined with over 1200 sets of historical case data from shipping accident reports and over 5000 sets of CFD thermal-fluid coupling and chemical reaction kinetics multiphysics simulation data to collect valid samples. Each sample included: the rate of change of temperature difference between the upper and lower cargo holds per unit time ΔT, the increase in hydrogen concentration ΔH2, and the time-series data of the ship's roll angle θ. These samples were used to train a four-layer BP neural network. The input layer had 3 nodes, the first hidden layer had 12 nodes (ReLU activation), the second hidden layer had 8 nodes (ReLU activation), and the output layer had 5 nodes (Softmax activation, outputting the probability distribution of five risk levels). After training, the model weight file was embedded into the multi-parameter fusion prediction module. In actual operation, when ΔT = 5.5℃ / h, ΔH2 = 0.8%vol / h, and θ = 12°, the model outputs a risk level of Level 3 (alarm), with a calculated warning time Tpredict of 35 minutes (coupling correction coefficient α = 0.8), meeting the 30-minute warning window requirement ahead of schedule. The system then triggers a Level 3 alarm response. When the multi-parameter fusion prediction module outputs a risk level of Level 2 (alarm), the measured hydrogen concentration is 0.3%vol. Based on the OR mechanism in the "dual-input logic gate," the system directly triggers a Level 2 warning, activating the top-mounted fan for active dilution (ventilation volume 3 times / hour). This demonstrates that the model prediction surpasses the hard threshold trigger, achieving "active prevention." If the risk further escalates to Level 3 (alarm), with the hydrogen concentration reaching 1.2%vol, the system immediately activates the nitrogen inerting system, injecting nitrogen at a standard flow rate of 50m³ / h. The oxygen sensor provides real-time feedback on the oxygen content inside the chamber, reducing it from 21%vol to 4.8%vol within 30 seconds, achieving the combustion suppression target of less than 5%vol. The system immediately switches to intermittent nitrogen replenishment mode to maintain stable oxygen levels. In extreme cases where the risk escalates to Level 4 (emergency) and the hydrogen concentration reaches 4.5% vol, the system simultaneously executes the following: increasing the nitrogen flow rate to 150 m³ / h; reducing the oxygen concentration to 1.8% vol (≤2% vol target) within 60 seconds; fully opening the explosion-proof pressure relief vents; cutting off all non-explosion-proof power supplies around the cargo hold; and activating the hatch cover locking mechanism. The entire process response time is less than 5 seconds.

[0047] The HBI intelligent safety control method and control system for cargo holds of transport ships of the present invention are essentially the same, and will not be described in detail here.

[0048] However, those skilled in the art should recognize that the above embodiments are only used to illustrate the present invention and are not intended to limit the present invention. Any changes or modifications to the above embodiments that are within the essential spirit of the present invention will fall within the scope of the claims of the present invention.

Claims

1. An intelligent safety control system for the cargo hold of an HBI transport ship, characterized in that, include: A multi-parameter sensor system collects real-time data on temperature, hydrogen concentration, oxygen concentration, and ship attitude in the upper and lower cargo holds. The multi-parameter fusion prediction module has a built-in real-time fusion prediction unit based on BP neural network. The real-time fusion prediction unit takes the temperature difference rate between the upper and lower layers of the cargo hold, the hydrogen concentration change rate, and the ship's roll angle as input parameters. After feature extraction by the double hidden layer, the Softmax layer outputs the probability distribution of five risk levels and calculates the warning time based on the extrapolation model of the coupling correction coefficient to ensure that the warning is given at least 30 minutes in advance. The tiered emergency response module receives the output signal from the multi-parameter fusion prediction module and executes tiered handling instructions according to preset logic.

2. The HBI intelligent safety control system for cargo holds of transport ships as described in claim 1, characterized in that: The multi-parameter sensor system includes temperature sensors respectively installed at the bottom and top of the cargo hold, hydrogen concentration sensors and oxygen sensors installed at the top of the cargo hold, and ship attitude sensors installed on the rigid structure in the middle of the cargo hold.

3. The HBI intelligent safety control system for cargo holds of transport ships as described in claim 2, characterized in that: The multi-parameter fusion prediction module uses a four-layer BP neural network with three nodes in the input layer. It receives the three-dimensional input feature vector X=[ΔT_norm,ΔH2_norm,θ_norm] after Z-score normalization. Where ΔT_norm is the normalized rate of temperature difference between the upper and lower layers, ΔH2_norm is the normalized rate of hydrogen concentration change, and θ_norm is the normalized ship roll angle; the first hidden layer has 12 nodes, uses the ReLU activation function, and outputs h. (1) =ReLU(W (1) ·X+b (1) ), where W (1) The weight matrix is ​​12×3, b (1) It is a 12×1 bias vector; The second hidden layer has 8 nodes, using the ReLU activation function, and outputs h. (2) =ReLU(W (2) ·h (1) +b (2) ), where W (2) The weight matrix is ​​8×12, b (2) The bias vector is 8×1; the output layer has 5 nodes, uses the Softmax activation function, and the output Risk_Level = Softmax(W (3) ·h (2) +b (3) ), where W (3) The weight matrix is ​​5×8, b (3) The bias vector is 5×1. Each element of the output vector corresponds to the predicted probability of the five risk levels. The sum of all elements is 1. The category with the highest probability is taken as the output risk level.

4. The HBI intelligent safety control system for cargo holds of transport ships as described in claim 1, characterized in that: The formula for calculating the warning time Tpredict is: In the formula, T predict Threshold is the warning time. critical The critical threshold is set to either 1%vol for hydrogen concentration or 100℃ for temperature. Current_Value is the current measured value, dX / dt is the parameter change rate, and α(ΔH2, θ) is the coupling correction coefficient.

5. The HBI intelligent safety control system for cargo holds of transport ships as described in claim 4, characterized in that: The coupling correction coefficient α is determined according to the following rules: when ΔH2 ≥ 1.0%vol / h, α = 0.8; when θ ≥ 15°, α = 0.7; when ΔH2 ≥ 1.0%vol / h and θ ≥ 15°, α = 0.6; and under normal operating conditions, α = 1.

0.

6. The HBI intelligent safety control system for cargo holds of transport ships as described in claim 5, characterized in that: The input-output logic decision matrix of the multi-parameter fusion prediction module is as follows: when ΔT < 3℃ / h, ΔH2 < 0.1%vol / h, and θ < 10°, output Level 0, system standby monitoring; when ΔT ≥ 3 and < 5℃ / h, ΔH2 < 0.5%vol / h, and θ < 10°, output Level 1, warning time ≥ 30min; when ΔT ≥ 5℃ / h, ΔH2 is between 0.5-1.0%vol / h, and θ < 10°, output Level 2, warning time ≥ 30min; when ΔT ≥ 5℃ / h or temperature ≥ 100℃, ΔH2 is between 1.0-4.0%vol / h, and θ ≥ 10°, output Level 3, warning time ≥ 30min; when temperature ≥ 150℃, ΔH2 ≥ 4.0%vol / h, and θ ≥ 15°, output Level 4.

7. The HBI intelligent safety control system for cargo holds of a transport ship as described in claim 6, characterized in that: The tiered emergency response module includes four levels of linkage units. Its triggering condition is determined by a logic gate composed of the risk level output by the multi-parameter fusion prediction module and the sensor's measured threshold. Specifically, it includes: Level 1 early warning unit: When the risk level is Level 1 and ΔT≥3℃ / h and H2≤0.5%vol, increase the sampling frequency of the multi-parameter sensor system to 1 time / 10 seconds; Level 2 warning unit: When the risk level is Level 2 and ΔT≥5℃ / h and H2 is between 0.5% and 1.0%vol, the forced ventilation at the top of the cargo hold will be automatically activated, with a ventilation rate of 3 air changes per hour; Level 3 alarm unit: When the risk level is Level 3 and H2 is between 1.0% and 4.0% vol or the temperature is ≥100℃, the nitrogen inerting system is activated, injecting nitrogen into the cargo hold at a standard flow rate of 50 m³ / h to reduce the oxygen volume content in the hold to below 5% within 30 seconds to suppress combustion. Afterward, nitrogen is intermittently replenished through closed-loop feedback from the oxygen sensor to maintain the oxygen supply. During this time, the explosion-proof pressure relief vents in the cargo hold remain closed. Level 4 Emergency Unit: When the risk level is Level 4 and H2 ≥ 4.0% vol or temperature ≥ 150℃, the following actions shall be performed simultaneously: increase the nitrogen injection rate to an enhanced flow rate of 150 m³ / h, reduce the oxygen volume content in the cabin to below 2% within 60 seconds to suppress the explosion, and perform a forced inerting cycle throughout the cabin; open all explosion-proof pressure relief vents in the cargo hold; cut off all non-explosion-proof power supplies around the cargo hold; and lock the cargo hold hatch covers.

8. The HBI intelligent safety control system for cargo holds of transport ships as described in claim 7, characterized in that: The response sequence of the hierarchical linkage emergency module is as follows: signal transmission delay <100ms, controller processing cycle <50ms, actuator mechanical response <3s, and total response time from signal acquisition to linkage execution <5 seconds.

9. The HBI intelligent safety control system for cargo holds of transport ships as described in claim 1, characterized in that: It also includes a redundant communication module and a data storage unit. The redundant communication module includes a CAN bus wired communication unit and a 5G wireless communication unit, with an end-to-end data latency of <100ms. The data storage unit is used to store no less than 6 months of continuous monitoring data.

10. The prevention and control method of the HBI intelligent safety control system for cargo holds of a transport ship as described in any one of claims 1-9, characterized in that, Includes the following steps: A. Real-time data collection of temperature gradient, hydrogen concentration, oxygen concentration, and ship attitude data within the cargo hold using a multi-parameter sensor system; B. Through a multi-parameter fusion prediction module with a built-in real-time fusion prediction unit based on a BP neural network, the temperature difference rate between the upper and lower layers of the cargo hold, the rate of change of hydrogen concentration, and the ship's roll angle are used as input parameters. After feature extraction by the double hidden layer, the Softmax layer outputs the probability distribution of five risk levels, and the warning time is calculated based on the extrapolation model of the coupling correction coefficient to ensure that the warning is given at least 30 minutes in advance. C. The hierarchical linkage emergency module receives the output signal of the multi-parameter fusion prediction module and executes hierarchical handling instructions according to preset logic.

Images