Intelligent emergency water injection safety system suitable for liquefied petroleum gas storage tank

Abstract

The present application relates to the technical field of petroleum gas storage tank safety, in particular to an intelligent emergency water injection safety system suitable for liquefied petroleum gas storage tank, comprising a physical data acquisition unit, a digital model mapping unit, a dynamic risk assessment unit and a water injection strategy optimization unit. The dynamic risk assessment unit of the present application is based on a nonlinear coupled equation set and an environmental correction mechanism, realizes accurate calculation of multidimensional safety factors and dynamic determination of risk levels, combines confidence interval analysis and multi-angle verification to ensure the comprehensiveness and reliability of risk assessment; the water injection strategy optimization unit relies on an improved reinforcement learning model, combines meta-learning initialization, adaptive network architecture and information entropy guided exploration strategy to generate the optimal water injection strategy and perform virtual verification, forms a closed-loop optimization mechanism, significantly improves the safety, response efficiency and adaptive ability of the system, and realizes a leapfrog upgrade from passive monitoring to active intelligent decision-making.

Description

Technical Field

[0001] This invention relates to the field of petroleum gas storage tank safety technology, specifically to an intelligent emergency water injection safety system suitable for liquefied petroleum gas storage tanks. Background Technology

[0002] Liquefied petroleum gas (LPG) storage tanks are containers used to store LPG. They are typically made of steel and possess excellent pressure and corrosion resistance. They are primarily used to store LPG produced during crude oil refining or natural gas processing. This gas can be liquefied at room temperature by pressurization and stored in the tanks. The structural design of the tanks ensures safety and stability, preventing accidents such as leaks and explosions. LPG storage tanks are widely used in chemical, energy, and transportation industries and are essential facilities for storing and transporting LPG.

[0003] However, conventional water injection systems have certain limitations in application. These limitations mainly manifest as fixed parameters and delayed updates, making it difficult for the model to accurately reflect changes in actual operating conditions and reducing the system's real-time response capability. At the same time, traditional risk assessment methods often rely on a single threshold judgment, lacking the ability to dynamically correct for environmental disturbances and failing to comprehensively consider the coupling relationship between multiple parameters, thus affecting the accuracy and comprehensiveness of risk prediction. In addition, learning models in water injection safety systems generally suffer from fixed network structures and inefficient exploration strategies, making them prone to overfitting and unable to adapt to the complex and ever-changing tank operating environment. They also suffer from slow convergence speed, affecting the system's optimization efficiency and decision-making speed, thus restricting the reliability and adaptability of intelligent emergency water injection safety systems in practical applications.

[0004] Based on this, the present invention provides an intelligent emergency water injection safety system suitable for liquefied petroleum gas storage tanks to solve the above-mentioned technical problems. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent emergency water injection safety system suitable for liquefied petroleum gas storage tanks. The digital model mapping unit of this invention constructs a high-fidelity physical coupling model through deep coupling of a parametric geometric model and a set of physical parameters, supporting state prediction and dynamic calibration, and improving the accuracy of virtual-to-real mapping. The dynamic risk assessment unit, based on a nonlinear coupled equation system and an environmental correction mechanism, achieves accurate calculation of multi-dimensional safety factors and dynamic determination of risk levels. Combined with confidence interval analysis and multi-angle verification, it ensures the comprehensiveness and reliability of risk assessment. The water injection strategy optimization unit relies on an improved reinforcement learning model, combined with meta-learning initialization, adaptive network architecture, and an information entropy-guided exploration strategy, to generate the optimal water injection strategy and perform virtual verification, forming a closed-loop optimization mechanism. This significantly improves the system's safety, response efficiency, and adaptability, achieving a leapfrog upgrade from passive monitoring to proactive intelligent decision-making.

[0006] To achieve the above objectives, the present invention provides the following technical solution: This invention provides an intelligent emergency water injection safety system suitable for liquefied petroleum gas storage tanks, comprising a physical data acquisition unit, a digital model mapping unit, a dynamic risk assessment unit, and a water injection strategy optimization unit, wherein: The physical data acquisition unit is used to acquire multiple sensor data from the liquefied petroleum gas storage tank in real time, and to preprocess the multiple sensor data to obtain the first dataset. The digital model mapping unit: transforms the liquefied petroleum gas storage tank into a parametric geometric model, assigns values ​​based on the physical properties and material characteristics of the liquefied petroleum gas storage tank to obtain a set of physical parameters, couples the set of physical parameters with the first dataset, and outputs a physical coupling model. The dynamic risk assessment unit: combines the physical coupling model with the first dataset to define parameter boundaries, obtains the safety factor, performs analysis based on the safety factor and simulates in the physical coupling model, outputs simulation results, constructs a risk assessment function and assessment matrix, and obtains the risk assessment results; The water injection strategy optimization unit: uses the risk assessment results to output a safety state vector, designs a reward function and a reinforcement learning model, outputs the optimal water injection strategy through the reinforcement learning model, verifies the optimal water injection strategy, and obtains a water injection instruction set.

[0007] The physical data acquisition unit includes a multimodal sensing module and a data preprocessing module, wherein: The multimodal sensing module integrates data from multiple sensors, aligns the data from multiple sensors on timestamps, and forms a synchronized dataset. The data preprocessing module cleans and removes outliers from the synchronized dataset, and then standardizes and normalizes the cleaned synchronized dataset to obtain the first dataset. The first dataset includes: pressure, temperature, and liquid level data of the liquefied petroleum gas storage tank body, as well as ambient wind speed, humidity, and gas concentration data.

[0008] The digital model mapping unit includes a geometric assignment module and a physical mapping module, wherein: The geometry assignment module performs topology extraction on the liquefied petroleum gas storage tank to obtain a parametric geometric model. It then assigns values ​​to the parametric geometric model using the material properties of the storage tank and calculates the relationship between the thermal expansion coefficient and temperature, as well as the relationship between liquefied petroleum gas and temperature and pressure, to obtain a set of physical parameters. The physical mapping module couples the flow field-temperature field, the tank wall stress-strain, and the wind speed-humidity to form a set of coupled equations. It then maps the first dataset to the assigned parameterized geometric model to obtain a physical coupling model. The physical coupling model is used to predict the future state and obtain the predicted value.

[0009] The dynamic risk assessment unit includes: a physical constraint module, a parameter analysis module, and a risk assessment module, wherein: The physical constraint module extracts influencing factors through the physical coupling model and the first dataset, calculates the safety factor based on the influencing factors and the physical parameters of the storage tank, and calibrates the safety factor. The multi-parameter analysis module: constructs a set of nonlinear coupled equations, obtains a coupling matrix, performs simulation in a physical coupling model, outputs simulation results, and calculates the risk level threshold based on the simulation results; The risk assessment module calculates and integrates the confidence intervals of the risk levels to obtain the assessment matrix, calculates the deviation between the predicted values ​​and the first dataset, calibrates the physical coupling model based on the deviation, and performs multi-angle verification in conjunction with the assessment matrix to obtain the risk assessment results.

[0010] The physical constraint module extracts influencing factors from the physical coupling model and the first dataset, calculates the safety factor based on the influencing factors and the physical parameters of the storage tank, and calibrates the safety factor, including: The influencing factors include pressure factor, temperature factor, and thermal resistance factor. The non-uniform distribution of liquefied petroleum gas molecules in the storage tank is simulated in the physical coupling model, and the pressure factor, temperature factor and thermal resistance factor are calculated. The ideal gas equation and the heat conduction equation are modified by using pressure factor, temperature factor and thermal resistance factor respectively to obtain pressure and temperature safety factors. The liquid level factor and liquid level safety factor are obtained by using the fluctuation frequency and amplitude of liquid level. The difference between the predicted value and the safety factor is calculated, and the physical coupling model is calibrated based on the deviation to obtain the calibrated safety factor.

[0011] The multi-parameter analysis module constructs a set of nonlinear coupled equations, analyzes the interactions between parameters, obtains the coupling matrix, performs simulations in a physical coupling model, outputs simulation results, and calculates the risk level threshold based on the simulation results. The specific operations are as follows: Based on the calibrated safety factor and the wind speed, humidity and gas concentration data of the first dataset, a set of nonlinear coupled equations is established. The nonlinear coupled equations are solved to obtain the coupling matrix and the disorder factor. A three-dimensional coupled equation set is constructed in the physical coupling model, and wind speed and humidity are used as boundary conditions for calculation to obtain simulation results and environmental correction coefficients; The maximum value of the calibrated safety factor is multiplied by the environmental correction factor and the disturbance factor to obtain the first risk level threshold; the average value of the calibrated safety factor is multiplied by the environmental correction factor to obtain the second risk level threshold; the minimum value of the calibrated safety factor is multiplied by the environmental correction factor to obtain the third risk level threshold.

[0012] The process involves constructing a three-dimensional coupled equation set within the physical coupling model, using wind speed and humidity as boundary conditions for calculation, to obtain simulation results and environmental correction coefficients. Specifically: The three-dimensional coupling equation set is as follows: Flow field:

[0013] Temperature field:

[0014] Concentration field:

[0015] In the formula: v is the velocity vector. Where is the dynamic viscosity, and g is the acceleration due to gravity. For specific heat capacity, Thermal conductivity, For heat source items, Where is the concentration and D is the diffusion coefficient; Embedding the coupling matrix into the system of equations, wind speed Solving for boundary conditions and using humidity After correcting the diffusion coefficient, we get: Gas diffusion path:

[0016] Temperature field on the surface of the storage tank:

[0017] Liquid level fluctuation:

[0018] In the formula: Where is the diffusion radius, For the calibrated temperature safety factor, The safety factor for the calibrated liquid level. This is the fluctuation coefficient.

[0019] The calculation steps for the environmental correction factor are as follows: Wind speed correction factor, humidity correction factor and concentration correction factor are calculated separately. The three types of correction factors are processed according to different scenarios to obtain the environmental correction factor.

[0020] The specific steps for obtaining the coupling matrix and the disorder factor are as follows: By embedding environmental data such as wind speed, humidity, and gas concentration into equations for pressure, temperature, and liquid level change rate, a coupling matrix and a four-dimensional coupling equation set are obtained. The test value is calculated by numerically solving the four-dimensional coupled equation system. If the test value is greater than 0, the disorder factor is set to 1; otherwise, the disorder factor is set to 0.

[0021] The water injection strategy optimization unit includes: a reinforcement learning module and a water injection strategy module, wherein: The reinforcement learning module integrates the risk assessment results with the calibrated safety coefficients to obtain a safety state vector, and designs immediate reward functions and long-term reward functions respectively. The water injection strategy module: improves the reinforcement learning model, trains the improved reinforcement learning model to obtain a trained reinforcement learning model, uses the trained reinforcement learning model to generate the optimal water injection strategy and verify it, and converts the verified optimal water injection strategy into a water injection instruction set.

[0022] Compared with the prior art, the beneficial effects of the present invention are: The digital model mapping unit of this invention constructs a high-fidelity physical coupling model through deep coupling of a parametric geometric model and a physical parameter set, supporting state prediction and dynamic calibration, and improving the accuracy of virtual-to-real mapping. The dynamic risk assessment unit, based on a nonlinear coupling equation system and an environmental correction mechanism, achieves accurate calculation of multi-dimensional safety coefficients and dynamic determination of risk levels. Combined with confidence interval analysis and multi-angle verification, it ensures the comprehensiveness and reliability of risk assessment. The water injection strategy optimization unit relies on an improved reinforcement learning model, combined with meta-learning initialization, adaptive network architecture, and information entropy-guided exploration strategy, to generate the optimal water injection strategy and perform virtual verification, forming a closed-loop optimization mechanism. This significantly improves the system's security, response efficiency, and adaptive capabilities, achieving a leapfrog upgrade from passive monitoring to proactive intelligent decision-making. Attached Figure Description

[0023] Figure 1 This is a system diagram of an intelligent emergency water injection safety system for liquefied petroleum gas storage tanks according to the present invention. Detailed Implementation

[0024] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention. Example

[0025] like Figure 1 As shown, this embodiment provides an intelligent emergency water injection safety system suitable for liquefied petroleum gas storage tanks. The present invention provides an intelligent emergency water injection safety system suitable for liquefied petroleum gas storage tanks, including a physical data acquisition unit, a digital model mapping unit, a dynamic risk assessment unit, and a water injection strategy optimization unit, wherein: The physical data acquisition unit is used to acquire multiple sensor data from the liquefied petroleum gas storage tank in real time, and to preprocess the multiple sensor data to obtain the first dataset. The digital model mapping unit: transforms the liquefied petroleum gas storage tank into a parametric geometric model, assigns values ​​based on the physical properties and material characteristics of the liquefied petroleum gas storage tank to obtain a set of physical parameters, couples the set of physical parameters with the first dataset, and outputs a physical coupling model. The dynamic risk assessment unit: combines the physical coupling model with the first dataset to define parameter boundaries, obtains the safety factor, performs analysis based on the safety factor and simulates in the physical coupling model, outputs simulation results, constructs a risk assessment function and assessment matrix, and obtains the risk assessment results; The water injection strategy optimization unit: uses the risk assessment results to output a safety state vector, designs a reward function and a reinforcement learning model, outputs the optimal water injection strategy through the reinforcement learning model, verifies the optimal water injection strategy, and obtains a water injection instruction set.

[0026] The physical data acquisition unit includes a multimodal sensing module and a data preprocessing module, wherein: The multimodal sensing module integrates data from multiple sensors, aligns the data from multiple sensors on timestamps, and forms a synchronized dataset. The data preprocessing module cleans and removes outliers from the synchronized dataset, and then standardizes and normalizes the cleaned synchronized dataset to obtain the first dataset. It should be noted that the specific operations for cleaning the synchronous dataset are as follows: cubic spline interpolation is used to fill in the missing data caused by sensor failure, and outliers are identified and removed based on the 3σ criterion. At the same time, a moving average filter is used to smooth the cleaned data. The above are all existing technologies, and will not be elaborated further in this embodiment.

[0027] The first dataset includes: pressure, temperature, and liquid level data of the liquefied petroleum gas storage tank body, as well as ambient wind speed, humidity, and gas concentration data.

[0028] The digital model mapping unit: transforms the liquefied petroleum gas (LPG) storage tank into a parametric geometric model, assigns values ​​based on the physical properties and material characteristics of the LPG storage tank to obtain a physical parameter set, couples the physical parameter set with the first dataset, and outputs a physical coupling model, wherein: The geometric assignment module performs topological extraction on the liquefied petroleum gas storage tank based on CAD drawings and point cloud data to obtain a parametric geometric model. It then assigns values ​​to the parametric geometric model using the material properties of the storage tank and calculates the relationship between the thermal expansion coefficient and temperature, as well as the relationship between liquefied petroleum gas and temperature and pressure, to obtain a set of physical parameters. At the same time, dimensional relationships are established, such as associating the tank height with the solvent, to enable automatic updates of the parametric geometric model when parameters are modified.

[0029] The equations relating the coefficient of thermal expansion to temperature, and the equations relating liquefied petroleum gas to temperature and pressure, are as follows:

[0030]

[0031] In the formula: The coefficient of thermal expansion at the reference temperature. This is the temperature coefficient, and its value depends on the material's inherent properties. The change in temperature For pressure, The gas constant is... Indicates molar mass; The physical mapping module couples the flow field-temperature field, the tank wall stress-strain, and the wind speed-humidity to form a set of coupled equations. It then maps the first dataset to the assigned parameterized geometric model to obtain a physical coupling model. The physical coupling model is used to predict the future state and obtain the predicted value.

[0032] In this embodiment: the stress tensor is:

[0033] In the formula: For elastic modulus, Poisson's ratio, For strain tensor, Kronecker delta function; The first dataset is mapped to the parameterized geometric model after assignment through the data interface, so as to realize the real-time update of the model state. In this embodiment, the real-time update of the model state can be predicted by the finite volume method, but this application does not limit it.

[0034] The dynamic risk assessment unit includes: a physical constraint module, a parameter analysis module, and a risk assessment module, wherein: The physical constraint module extracts influencing factors through the physical coupling model and the first dataset, calculates the safety factor based on the influencing factors and the physical parameters of the storage tank, and calibrates the safety factor. As one specific implementation method, the influencing factors include pressure factor, temperature factor, and thermal resistance factor; The non-uniform distribution of liquefied petroleum gas molecules in the storage tank is simulated in the physical coupling model, and the pressure factor, temperature factor and thermal resistance factor are calculated. Utilizing stress factors Temperature factor and thermal resistance factor By modifying the ideal gas equation and the heat conduction equation respectively, pressure and temperature safety factors are obtained. Then, by analyzing the frequency and amplitude of liquid level fluctuations, the liquid level factor and liquid level safety factor are derived. The stress factor is:

[0035] thermal resistance factor

[0036] In the formula: Where N is the local volume and N is the number of molecules. Molecular mass For molecular velocity, It is the normal vector. For surface area, For time step, For effective thermal conductivity, The thickness of the lattice layer. For heat transfer area; The revised ideal gas equation and heat conduction equation are as follows:

[0037]

[0038] In the formula: It is the saturated vapor pressure. For operating pressure, Specific heat capacity of the tank wall material; The temperature distribution of the tank wall was obtained through numerical solution. Then, the temperature safety factor is calculated:

[0039] The safety factor for liquid level is:

[0040] In the formula: This is the median value of the temperature safety threshold. This represents the maximum value of the temperature safety threshold. The minimum value of the temperature safety threshold. For the liquid level safety threshold, This is the minimum permissible liquid level.

[0041] The difference between the predicted value and the safety factor is calculated, and the physical coupling model is calibrated based on the deviation. If the deviation exceeds 5%, the model parameter calibration is triggered, and the molecular collision frequency, lattice thermal conductivity, etc. are dynamically adjusted to keep the model prediction value consistent with the real-time data, thus obtaining the calibrated safety factor.

[0042] The multi-parameter analysis module: constructs a set of nonlinear coupled equations, obtains a coupling matrix, performs simulation in a physical coupling model, outputs simulation results, and calculates the risk level threshold based on the simulation results; For example, the specific operation is as follows: Based on the calibrated safety factor and the wind speed, humidity and gas concentration data of the first dataset, a set of nonlinear coupled equations is established. The nonlinear coupled equations are solved to obtain the coupling matrix and the disorder factor. The specific steps for obtaining the coupling matrix and the disorder factor are as follows: By embedding environmental data such as wind speed, humidity, and gas concentration into equations for pressure, temperature, and liquid level change rate, a coupling matrix and a four-dimensional coupling equation set are obtained. Pressure change rate equation:

[0043] Equation for the rate of change of temperature:

[0044] Equation for rate of change of liquid level:

[0045] In the formula: , This is the pressure-temperature coupling coefficient, with values ​​of 0.8 and 0.3. The wind and humidity interference coefficient is set to 0.05. , This is the temperature-liquid level coupling coefficient. The concentration influence coefficient is set to 0.05. , The level-pressure coupling coefficient is... This is the wind speed fluctuation coefficient, with a value of 0.1; It should be added that the temperature-liquid level coupling coefficient is derived based on the experimental data of liquid level fluctuation and the coupling of the fluid dynamics continuity equation and the energy equation, while the liquid level-pressure coupling coefficient needs to be analyzed in conjunction with the flow field-liquid level coupling model, wind tunnel experiments and liquid level fluctuation spectrum.

[0046] The test value is calculated by numerically solving the four-dimensional coupled equation system. If the test value is greater than 0, the disorder factor is set to 1; otherwise, the disorder factor is set to 0.

[0047] The formula for calculating the test value is:

[0048] In the formula: = , is a four-dimensional vector; The process involves constructing a three-dimensional coupled equation set within the physical coupling model, using wind speed and humidity as boundary conditions for calculation, to obtain simulation results and environmental correction coefficients. Specifically: The three-dimensional coupling equation set is as follows: Flow field:

[0049] Temperature field:

[0050] Concentration field:

[0051] In the formula: v is the velocity vector. Where is the dynamic viscosity, and g is the acceleration due to gravity. For specific heat capacity, Thermal conductivity, For heat source items, Where is the concentration and D is the diffusion coefficient; Embedding the coupling matrix into the system of equations, wind speed Solving for boundary conditions and using humidity After correcting the diffusion coefficient, we get: Gas diffusion path:

[0052] Temperature field on the surface of the storage tank:

[0053] Liquid level fluctuation:

[0054] In the formula: Where is the diffusion radius, For the calibrated temperature safety factor, The safety factor for the calibrated liquid level. This is the fluctuation coefficient.

[0055] The calculation steps for the environmental correction factor are as follows: Calculate the wind speed correction factor separately. Humidity correction factor and concentration correction factor Based on different scenarios, the three types of correction coefficients are processed to obtain the environmental correction coefficients, namely:

[0056]

[0057]

[0058] In the formula: The wind speed sensitivity coefficient is set to 0.02. The humidity sensitivity coefficient has a value of 0.01. The concentration sensitivity coefficient is 0.05.

[0059] When environmental parameters are independent of each other, the environmental correction factor ; When facing a complex environment, that is... ; In the formula: For example, the pressure-temperature coupling coefficient. For nonlinear functions, such as , This is the magnification factor, with a value of 0.2.

[0060] The first risk level threshold is obtained by multiplying the maximum value of the calibrated safety factor by the environmental correction factor and the disturbance factor; the second risk level threshold is obtained by multiplying the average value of the calibrated safety factor by the environmental correction factor; and the third risk level threshold is obtained by multiplying the minimum value of the calibrated safety factor by the environmental correction factor.

[0061] The risk thresholds of multiple levels are compared in turn. If the first risk level threshold is greater than 0.8, it is directly regarded as high risk and requires immediate emergency intervention (such as starting a large-volume water injection). If the first risk level threshold is less than 0.8, but the second risk level threshold is between 0.5 and 0.8, it is considered a medium risk. If the third-level threshold is ≤0.5, it is considered low risk; The risk assessment module calculates and integrates the confidence intervals of the risk levels to obtain the assessment matrix, calculates the deviation between the predicted values ​​and the first dataset, calibrates the physical coupling model based on the deviation, and performs multi-angle verification in conjunction with the assessment matrix to obtain the risk assessment results.

[0062] The confidence intervals for risk levels are calculated and integrated, which involves random sampling of the safety factor 1000 times, statistical analysis of the risk level distribution, and determination of the risk level judgment result under the 95% confidence interval. The formula for calculating the deviation between the predicted value and the first dataset is:

[0063] In the formula: , , These are predicted values ​​for pressure, temperature, and liquid level. , , These are the actual values ​​of pressure, temperature, and liquid level.

[0064] When D is greater than 0.05, the calibration of the physical coupling model is triggered. The multi-angle verification based on the evaluation matrix is ​​as follows: if any of the predicted values ​​reaches 80% of the safety limit, it is directly judged as high risk; The water injection strategy optimization unit includes: a reinforcement learning module and a water injection strategy module, wherein: The reinforcement learning module integrates the risk assessment results with the calibrated safety coefficients to obtain a safety state vector, and designs immediate reward functions and long-term reward functions respectively. The instant reward function is:

[0065] The long-term reward function is:

[0066] In the formula: , , The weights for each safety factor are 0.5, 0.3, and 0.2. Inject a correction factor into the energy. The amount of energy injected. , The safety factor before and after the action is performed; The water injection strategy module: improves the reinforcement learning model, trains the improved reinforcement learning model to obtain a trained reinforcement learning model, uses the trained reinforcement learning model to generate the optimal water injection strategy and verify it, and converts the verified optimal water injection strategy into a water injection instruction set.

[0067] The reinforcement learning model is improved as follows: Model-independent meta-learning algorithm generates initial network parameters This allows the model to quickly adapt to new working conditions with a small number of samples. The meta-learning objective function is: ,in, Let N be the loss function for the i-th working condition, and N be the number of working conditions.

[0068] The network layer and number of neurons are dynamically adjusted based on four-dimensional vectors. If all values ​​in the four-dimensional vector are close to 1, the network is automatically expanded to 6 layers, and the number of neurons in each layer increases in powers of 2. If only one value is close to 1, the network is shrunk to 4 layers to avoid overfitting.

[0069] A cross-layer parameter sharing module is introduced between multi-layer networks, and an attention mechanism is used to dynamically allocate the attention different layers pay to state variables. ,in, Let W and b be the weights of the four-dimensional vector at the i-th layer, and let W and b be the training parameters. Based on the current strategy (S) Calculate the information entropy of the action: When the information entropy is greater than 0.8, the information increment maximization strategy is adopted to prioritize the exploration of high information entropy regions; when it is less than 0.8, the strategy is switched to safety boundary exploration to focus on exploring actions close to the safety threshold. Constructing a three-dimensional reward function:

[0070] In the formula: = ; = , for , , , , where is the weighting coefficient, and the values ​​are 0.6, 0.3, and 0.1.

[0071] Meanwhile, massive amounts of synthetic data are generated in a virtual environment using a physical coupling model to accelerate the training of reinforcement learning models. The synthetic data must meet physical consistency constraints and be validated in conjunction with the physical coupling model to train the reinforcement learning model.

[0072] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0073] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. An intelligent emergency water injection safety system suitable for liquefied petroleum gas storage tanks, characterized in that, It includes a physical data acquisition unit, a digital model mapping unit, a dynamic risk assessment unit, and a water injection strategy optimization unit, wherein: The physical data acquisition unit is used to acquire multiple sensor data from the liquefied petroleum gas storage tank in real time, and to preprocess the multiple sensor data to obtain the first dataset. The digital model mapping unit: transforms the liquefied petroleum gas storage tank into a parametric geometric model, assigns values ​​based on the physical properties and material characteristics of the liquefied petroleum gas storage tank to obtain a set of physical parameters, couples the set of physical parameters with the first dataset, and outputs a physical coupling model. The dynamic risk assessment unit: combines the physical coupling model with the first dataset to define parameter boundaries, obtains the safety factor, performs analysis based on the safety factor and simulates in the physical coupling model, outputs simulation results, constructs a risk assessment function and assessment matrix, and obtains the risk assessment results; The water injection strategy optimization unit: uses the risk assessment results to output a safety state vector, designs a reward function and a reinforcement learning model, outputs the optimal water injection strategy through the reinforcement learning model, verifies the optimal water injection strategy, and obtains a water injection instruction set.

2. The intelligent emergency water injection safety system for liquefied petroleum gas storage tanks according to claim 1, characterized in that, The physical data acquisition unit includes a multimodal sensing module and a data preprocessing module, wherein: The multimodal sensing module integrates data from multiple sensors, aligns the data from multiple sensors on timestamps, and forms a synchronized dataset. The data preprocessing module cleans and removes outliers from the synchronized dataset, and then standardizes and normalizes the cleaned synchronized dataset to obtain the first dataset. The first dataset includes: pressure, temperature, and liquid level data of the liquefied petroleum gas storage tank body, as well as ambient wind speed, humidity, and gas concentration data.

3. The intelligent emergency water injection safety system for liquefied petroleum gas storage tanks according to claim 1, characterized in that, The digital model mapping unit includes a geometric assignment module and a physical mapping module, wherein: The geometry assignment module performs topology extraction on the liquefied petroleum gas storage tank to obtain a parametric geometric model. It then assigns values ​​to the parametric geometric model using the material properties of the storage tank and calculates the relationship between the thermal expansion coefficient and temperature, as well as the relationship between liquefied petroleum gas and temperature and pressure, to obtain a set of physical parameters. The physical mapping module couples the flow field-temperature field, the tank wall stress-strain, and the wind speed-humidity to form a set of coupled equations. It then maps the first dataset to the assigned parameterized geometric model to obtain a physical coupling model. The physical coupling model is used to predict the future state and obtain the predicted value.

4. The intelligent emergency water injection safety system for liquefied petroleum gas storage tanks according to claim 1, characterized in that, The dynamic risk assessment unit includes: a physical constraint module, a parameter analysis module, and a risk assessment module, wherein: The physical constraint module extracts influencing factors through the physical coupling model and the first dataset, calculates the safety factor based on the influencing factors and the physical parameters of the storage tank, and calibrates the safety factor. The multi-parameter analysis module: constructs a set of nonlinear coupled equations, obtains a coupling matrix, performs simulation in a physical coupling model, outputs simulation results, and calculates the risk level threshold based on the simulation results; The risk assessment module calculates and integrates the confidence intervals of the risk levels to obtain the assessment matrix, calculates the deviation between the predicted values ​​and the first dataset, calibrates the physical coupling model based on the deviation, and performs multi-angle verification in conjunction with the assessment matrix to obtain the risk assessment results.

5. The intelligent emergency water injection safety system for liquefied petroleum gas storage tanks according to claim 4, characterized in that, The physical constraint module extracts influencing factors from the physical coupling model and the first dataset, calculates the safety factor based on the influencing factors and the physical parameters of the storage tank, and calibrates the safety factor, including: The influencing factors include pressure factor, temperature factor, and thermal resistance factor. The non-uniform distribution of liquefied petroleum gas molecules in the storage tank is simulated in the physical coupling model, and the pressure factor, temperature factor and thermal resistance factor are calculated. The ideal gas equation and the heat conduction equation are modified by using pressure factor, temperature factor and thermal resistance factor respectively to obtain pressure and temperature safety factors. The liquid level factor and liquid level safety factor are obtained by using the fluctuation frequency and amplitude of liquid level. The difference between the predicted value and the safety factor is calculated, and the physical coupling model is calibrated based on the deviation to obtain the calibrated safety factor.

6. The intelligent emergency water injection safety system for liquefied petroleum gas storage tanks according to claim 4, characterized in that, The multi-parameter analysis module constructs a set of nonlinear coupled equations, analyzes the interactions between parameters, obtains the coupling matrix, performs simulations in a physical coupling model, outputs simulation results, and calculates the risk level threshold based on the simulation results. The specific operations are as follows: Based on the calibrated safety factor and the wind speed, humidity and gas concentration data of the first dataset, a set of nonlinear coupled equations is established. The nonlinear coupled equations are solved to obtain the coupling matrix and the disorder factor. A three-dimensional coupled equation set is constructed in the physical coupling model, and wind speed and humidity are used as boundary conditions for calculation to obtain simulation results and environmental correction coefficients; The first risk level threshold is obtained by multiplying the maximum value of the calibrated safety factor by the environmental correction factor and the disturbance factor; the second risk level threshold is obtained by multiplying the average value of the calibrated safety factor by the environmental correction factor; and the third risk level threshold is obtained by multiplying the minimum value of the calibrated safety factor by the environmental correction factor.

7. The intelligent emergency water injection safety system for liquefied petroleum gas storage tanks according to claim 6, characterized in that, The process involves constructing a three-dimensional coupled equation set within the physical coupling model, using wind speed and humidity as boundary conditions for calculation, to obtain simulation results and environmental correction coefficients. Specifically: The three-dimensional coupling equation set is as follows: Flow field: Temperature field: Concentration field: In the formula: v is the velocity vector. Where is the dynamic viscosity, and g is the acceleration due to gravity. For specific heat capacity, Thermal conductivity, For heat source items, Where is the concentration and D is the diffusion coefficient; Embedding the coupling matrix into the system of equations, wind speed Solving for boundary conditions and using humidity After correcting the diffusion coefficient, we get: Gas diffusion path: Temperature field on the surface of the storage tank: Liquid level fluctuation: In the formula: Where is the diffusion radius, For the calibrated temperature safety factor, The safety factor for the calibrated liquid level. This is the fluctuation coefficient.

8. The intelligent emergency water injection safety system for liquefied petroleum gas storage tanks according to claim 6, characterized in that, The calculation steps for the environmental correction factor are as follows: Wind speed correction factor, humidity correction factor and concentration correction factor are calculated separately. The three types of correction factors are processed according to different scenarios to obtain the environmental correction factor.

9. The intelligent emergency water injection safety system for liquefied petroleum gas storage tanks according to claim 6, characterized in that, The specific steps for obtaining the coupling matrix and the disorder factor are as follows: By embedding environmental data such as wind speed, humidity, and gas concentration into equations for pressure, temperature, and liquid level change rate, a coupling matrix and a four-dimensional coupling equation set are obtained. The test value is calculated by numerically solving the four-dimensional coupled equation system. If the test value is greater than 0, the disorder factor is set to 1; otherwise, the disorder factor is set to 0.

10. The intelligent emergency water injection safety system for liquefied petroleum gas storage tanks according to claim 1, characterized in that, The water injection strategy optimization unit includes: a reinforcement learning module and a water injection strategy module, wherein: The reinforcement learning module integrates the risk assessment results with the calibrated safety coefficients to obtain a safety state vector, and designs immediate reward functions and long-term reward functions respectively. The water injection strategy module: improves the reinforcement learning model, trains the improved reinforcement learning model to obtain a trained reinforcement learning model, uses the trained reinforcement learning model to generate the optimal water injection strategy and verify it, and converts the verified optimal water injection strategy into a water injection instruction set.

Images