Liquid hydrogen storage tank pressure prediction method, device and system

Abstract

The invention discloses a liquid hydrogen storage tank pressure prediction method, device and system, and the method comprises the steps: obtaining process parameters of a liquid hydrogen storage tank, and carrying out the preprocessing, and obtaining input parameters; and inputting the input parameters into the trained liquid hydrogen storage tank pressure prediction model, and outputting an in-tank pressure prediction value of the liquid hydrogen storage tank after the predicted storage duration. According to the method, the adverse effect on the generalization performance of the neural network due to too few data samples is reduced, and the precision of predicting the pressure of the liquid hydrogen storage tank is effectively improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of liquid hydrogen storage tanks, and in particular relates to a method, device and system for predicting the pressure of a liquid hydrogen storage tank. Background Art

[0002] In the 1970s, NASA began experimental research on the pressure changes inside liquid hydrogen tanks. However, so far, due to the high cost and high risk of self-pressurization tests of liquid hydrogen tanks, there are only five public tests, of which the typical ones are tests on MHTB tanks and K-site tanks. At present, the pressure prediction model of liquid hydrogen tanks usually adopts thermodynamic models and CFD models. Thermodynamic models need to make assumptions about the two-phase state inside the liquid hydrogen tank and simplify the heat transfer process inside and outside the tank. Since the limited space natural convection phenomenon in the two-phase flow zone is simplified to the limited space large plate natural convection process in thermodynamics, its natural convection heat transfer coefficient is determined by empirical formula, which leads to a certain error between the model prediction results and the actual results. In addition, the thermodynamic model with fewer assumptions has a higher complexity, and the compilation and calculation process is more complicated. In contrast, the CFD model can capture the flow and heat transfer details of the gas phase, liquid phase and two-phase region in the tank in more detail. The multiphase flow model VOF is usually combined with the evaporation and condensation Lee phase change model for simulation calculation. However, since the effect of pressure change on the heat and mass transfer exchange rate of the phase change process and the adiabatic assumption of the inner wall of the tank are not considered, the error gradually increases after a period of calculation. In addition, the CFD model requires a lot of time and resources during the calculation process, and it is difficult to provide accurate pressure prediction values ​​in a short time. In summary, the current use of thermodynamic models and CFD models to predict the pressure of liquid hydrogen storage tanks has the problem of insufficient accuracy and the problem of too few experimental data samples limiting the generalization ability of neural networks. Summary of the invention

[0003] In order to solve the deficiencies in the prior art, the present invention provides a method, device and system for predicting the pressure of a liquid hydrogen storage tank, which reduces the adverse effect of too few data samples on the generalization performance of the neural network and effectively improves the accuracy of the prediction of the pressure of the liquid hydrogen storage tank.

[0004] In order to achieve the above object, the technical solution adopted by the present invention is:

[0005] In a first aspect, a method for predicting the pressure of a liquid hydrogen storage tank is provided, comprising: obtaining process parameters of the liquid hydrogen storage tank and performing preprocessing to obtain input parameters; inputting the input parameters into a trained liquid hydrogen storage tank pressure prediction model, and outputting a predicted value of the pressure inside the liquid hydrogen storage tank after an expected storage time.

[0006] Furthermore, the process parameters of the liquid hydrogen storage tank include: the gas phase pressure inside the liquid hydrogen storage tank at the current moment , Liquid hydrogen temperature , gas hydrogen temperature , Liquid hydrogen volume , External ambient temperature , Total volume of storage tank , Tank internal surface area , Estimated storage time .

[0007] Furthermore, the pretreatment includes: calculating the ratio of the inner surface area of ​​the tank to the total volume of the tank according to the process parameters of the liquid hydrogen tank , Total heat leakage of liquid hydrogen storage tank , filling rate of liquid hydrogen storage tank , and the ratio of the internal surface area of ​​the tank to the total volume of the tank , Total heat leakage of liquid hydrogen storage tank , filling rate of liquid hydrogen storage tank 、The current gas phase pressure inside the liquid hydrogen storage tank and estimated storage duration as an input parameter.

[0008] Furthermore, the liquid hydrogen storage tank pressure prediction model is constructed based on the BPNN back-propagation neural network model, including: an input layer, multiple hidden layers and an output layer, the input layer includes five input layer nodes, and the output layer includes one output layer node.

[0009] Furthermore, the training method of the liquid hydrogen storage tank pressure prediction model includes: first using the liquid hydrogen storage tank thermodynamic partition model to generate a large sample data set, using the number of hidden layers HLN, the number of neurons in each hidden layer HLS, and the initial learning rate ILR as hyperparameters, and based on the Bayesian optimization algorithm, training and verifying the neural network model defined by different hyperparameter combinations to obtain a pre-trained model; then using a small sample data set based on transfer learning to train the pre-trained model, and obtaining the liquid hydrogen storage tank pressure prediction model by fine-tuning the parameters.

[0010] Furthermore, the thermodynamic zoning model of the liquid hydrogen storage tank adopts the following assumptions: (1) The liquid hydrogen storage tank is divided into three regions: the superheated region, the saturated region and the supercooled region. Each region is a control body with uniform parameters inside the control body; (2) The two-phase saturated region in the superheated region, the saturated region and the supercooled region is a massless interface. The physical properties of hydrogen inside the interface are saturated and the pressure is equal to the saturated pressure; (3) The total heat leakage of the liquid hydrogen storage tank is composed of the heat leakage of the wall insulation structure, the supporting structure and the pipeline, and the changes of the heat leakage density with time and space are not considered.

[0011] Furthermore, Adam is selected as the learning algorithm of the pre-trained model, and the MSE function is selected as the loss function; the cosine annealing algorithm is used to adjust the learning rate of the model under each set of hyperparameter combinations to accelerate the convergence speed of model training; and a five-fold cross-validation training method is used to achieve effective estimation of model performance and generalization ability.

[0012] Furthermore, the parameter fine-tuning includes: training the pre-trained model using a model-based parameter fine-tuning method, specifically: fine-tuning the weights and biases of the input layer, the first hidden layer and the output layer, and freezing the remaining parameters; adopting an exhaustive idea, further comparing the fine-tuning strategies of the input layer, the first hidden layer and the output layer using a five-fold cross-validation method, and finally obtaining the liquid hydrogen storage tank pressure prediction model.

[0013] In a second aspect, a liquid hydrogen storage tank pressure prediction device is provided, including: a data acquisition module for acquiring process parameters of the liquid hydrogen storage tank; a pressure prediction module for inputting the acquired process parameters of the liquid hydrogen storage tank into a trained liquid hydrogen storage tank pressure prediction model, and outputting a predicted value of the internal pressure of the liquid hydrogen storage tank after an expected storage time.

[0014] In a third aspect, a liquid hydrogen storage tank pressure prediction system is provided, comprising: a storage medium and a processor; the storage medium is used to store instructions; the processor is used to operate according to the instructions to execute the liquid hydrogen storage tank pressure prediction method described in the first aspect.

[0015] Compared with the prior art, the beneficial effects achieved by the present invention are as follows: the present invention inputs the collected process parameters of the liquid hydrogen storage tank into a trained liquid hydrogen storage tank pressure prediction model, and outputs a predicted value of the internal pressure of the liquid hydrogen storage tank after the expected storage time, thereby reducing the adverse effect of too few data samples on the generalization performance of the neural network and effectively improving the accuracy of the liquid hydrogen storage tank pressure prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a schematic diagram of the model structure of a pre-trained model in a liquid hydrogen storage tank pressure prediction method provided in an embodiment of the present invention;

[0017] Figure 2 Schematic diagram of the calculation process of the thermodynamic partition model of the liquid hydrogen storage tank in the embodiment of the present invention;

[0018] Figure 3 is a schematic diagram of a parameter fine-tuning strategy based on a transfer learning method in an embodiment of the present invention;

[0019] Figure 4 It is a schematic diagram of the method of using the liquid hydrogen storage tank pressure prediction model in an embodiment of the present invention. DETAILED DESCRIPTION

[0020] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0021] Embodiment 1

[0022] A method for predicting the pressure of a liquid hydrogen storage tank comprises: obtaining process parameters of the liquid hydrogen storage tank and performing preprocessing to obtain input parameters; inputting the input parameters into a trained liquid hydrogen storage tank pressure prediction model, and outputting a predicted value of the pressure inside the liquid hydrogen storage tank after an expected storage time.

[0023] like Figure 1~Figure 4 As shown, the present invention uses a transfer learning method to train a liquid hydrogen storage tank pressure prediction model based on a BPNN back propagation neural network model using a small amount of test data. , Liquid hydrogen temperature , gas hydrogen temperature , Liquid hydrogen volume , External ambient temperature , Total volume of storage tank , Tank internal surface area , Estimated storage time In the case of liquid hydrogen storage tank pressure prediction model, the predicted value of the liquid hydrogen storage tank pressure after the expected storage time is obtained. .

[0024] 1. Data collection and preprocessing

[0025] 1.1. Fill the liquid hydrogen storage tank, let it stand for a while after filling, and collect process parameters, including: the gas phase pressure inside the liquid hydrogen storage tank at the current moment , Liquid hydrogen temperature , gas hydrogen temperature , Liquid hydrogen volume , External ambient temperature , Total volume of storage tank , Tank internal surface area , Estimated storage time wait.

[0026] 1.2. Obtain the thermal conductivity, natural convection heat transfer coefficient, thermal radiation coefficient, thickness and other parameters of the liquid hydrogen storage tank structure, and calculate the total heat leakage based on the temperature difference between the liquid phase, gas phase and the external environment temperature. :

[0027]

[0028]

[0029]

[0030]

[0031]

[0032]

[0033]

[0034]

[0035]

[0036] in, , are the heat flux densities between the liquid phase, gas phase and wall, ; , are the heat transfer areas of the liquid phase and the gas phase, respectively. ; is the total heat flux density; , and They are radiation heat flux density, gas convection heat transfer density and solid conduction heat flux density, respectively. ; is the radiation thermal resistance, is the convective heat transfer coefficient, is the solid thermal conductivity, ; is the outside temperature of the tank, is the temperature inside the tank, For the aluminum foil reflective layer, . is the residual gas pressure, is the residual gas retention coefficient. , is the gas molar mass. Usually, for the residual gas in the vacuum chamber at room temperature, . For nylon mesh material, , It is the relative density of the gap between the insulation material and the solid material, usually taken as 0.02. is the actual thickness between the reflective layers, is the thermal conductivity of the spacer material.

[0037] 1.3. Use REFPROP V9.1 software to calculate the liquid hydrogen phase parameters at the current internal pressure and liquid hydrogen temperature of the liquid hydrogen storage tank to obtain the mass of the liquid hydrogen and the filling rate of the liquid hydrogen tank :

[0038]

[0039]

[0040] in, is the mass of liquid hydrogen, ; is the volume of liquid hydrogen, ; is the density of liquid hydrogen, ; is the current filling rate; is the total volume of the tank, .

[0041] 1.4. The ratio of the internal surface area of ​​the tank to the total volume of the tank , Total heat leakage of liquid hydrogen storage tank , filling rate of liquid hydrogen storage tank 、The current gas phase pressure inside the liquid hydrogen storage tank and estimated storage duration as an input parameter.

[0042] 2. Liquid hydrogen storage tank pressure prediction

[0043] 2.1. Constructing a liquid hydrogen storage tank pressure prediction model

[0044] A liquid hydrogen storage tank pressure prediction model is constructed based on the BPNN back propagation neural network model, including: an input layer, multiple hidden layers and an output layer, and five input layer nodes, one output layer node and multiple hidden layer nodes are set.

[0045] 2.2. The training method of the liquid hydrogen storage tank pressure prediction model includes:

[0046] 2.2.1. First, a large sample data set is generated using the thermodynamic partition model of a liquid hydrogen storage tank. The number of hidden layers HLN, the number of neurons in each hidden layer HLS, and the initial learning rate ILR are used as hyperparameters. Based on the Bayesian optimization algorithm, the neural network model defined by different hyperparameter combinations is trained and verified to obtain a pre-trained model.

[0047] Obtain the sample data required for BPNN neural network model training. The thermodynamic partition model method of liquid hydrogen storage tank is used to obtain a large sample model data set. The thermodynamic partition model assumes the following: (1) The liquid hydrogen storage tank is divided into three regions: superheated zone, saturated zone and supercooled zone. Each region is a control body with uniform parameters inside the control body; (2) The two-phase saturated zone in the superheated zone, saturated zone and supercooled zone is a massless interface. The physical properties of hydrogen inside the interface are saturated and the pressure is equal to the saturated pressure; (3) The total heat leakage of the liquid hydrogen storage tank is composed of the heat leakage of the wall insulation structure, the support structure and the pipeline, and the changes of the heat leakage density with time and space are not considered.

[0048] According to the thermodynamic partition theory, the fluid inside the liquid hydrogen storage tank can be described by the mass control equation and the energy control equation, and solved by Xian's discrete method in MATLAB scientific computing software. During the solution process, some physical property parameters of liquid hydrogen and gas hydrogen can be called by REFPROP V9.1 software in MATLAB. The specific steps are as follows: Figure 2 As shown;

[0049]

[0050]

[0051]

[0052]

[0053]

[0054]

[0055]

[0056]

[0057] in, , The density of gaseous hydrogen, the density of liquid hydrogen, ; , The volume of gaseous hydrogen and the volume of liquid hydrogen are respectively: ; , , They are gas hydrogen specific enthalpy, liquid hydrogen specific enthalpy, saturated gas hydrogen specific enthalpy, ; , They are the gas-interface heat transfer area, the liquid-interface heat transfer area, ; is the evaporation / condensation rate of liquid hydrogen, with evaporation as positive, ; is the saturated hydrogen temperature, K; , They are saturated gas hydrogen pressure and gas hydrogen pressure, Pa; is the volume fraction of gaseous hydrogen; , They are the gas-interface convective heat transfer coefficient, the liquid-interface convective heat transfer coefficient, ;

[0058] In the calculation process, the pressure threshold and maximum filling rate To calculate the termination conditions, the gas phase space pressure in the liquid hydrogen storage tank at different times can be obtained , Liquid hydrogen temperature , gas hydrogen temperature , Liquid hydrogen volume Total heat leakage Over time of changes.

[0059] Compared with other thermodynamic models, such as saturated homogeneous model, Russian model, boundary layer model, etc., the thermodynamic partition model has higher pressure prediction accuracy, moderate model compilation difficulty and calculation time. The model has high availability and credibility, and has been widely used in the pressure prediction of cryogenic fluid storage tanks such as LNG tanks, liquid nitrogen, and liquid hydrogen. The core idea of ​​the thermodynamic partition model is to divide the fluid inside the cryogenic tank into three phases: liquid phase, gas phase, and two-phase region. The thermodynamic behavior between the tank structure and the cryogenic fluid is simplified to the second-class boundary condition, that is, it is assumed that the cryogenic fluid inside the tank and the tank structure transfer heat through a fixed heat flux density. At the same time, the two-phase interface area, as the most significant area for heat and mass transfer between gas and liquid fluids, is simplified to a layer of massless interface. In the gas phase and liquid phase, the natural convection heat transfer phenomenon on the interface is simplified to the natural convection heat transfer phenomenon in a limited space. The heat and mass transfer exchange between the gas and liquid phases can be calculated by the relevant empirical formula.

[0060] Initial gas phase pressure , the ratio of the tank surface area to the total volume , initial filling rate Total heat leakage , Storage time Characterized by different moments The data is organized into a large sample dataset with five features and one label, which is then normalized for model training.

[0061] like Figure 1 As shown in the figure, a BP neural network model is constructed. The input layer has five input features and the output layer has one output feature. The number of hidden layers HLN, the number of neurons in each hidden layer HLS, and the initial learning rate ILR are used as hyperparameters. The Bayesian optimization method is used under the OptunaAPI architecture to determine the hyperparameters of the liquid hydrogen storage tank pressure prediction pre-training model in the python program. The neural network models defined by different hyperparameter combinations are trained and verified to obtain the pre-trained model.

[0062] In the Pytorch framework, the learning algorithm of the pre-trained model is Adam, and the loss function is the MSE function. In order to accelerate the convergence of the training process, the cosine annealing algorithm is used to adjust the learning rate of the model under each set of hyperparameter combinations to speed up the convergence of model training. The five-fold cross-validation training method is used to achieve effective estimation of model performance and generalization ability.

[0063] 2.2.2. Use the small sample data set based on transfer learning to train the pre-trained model, and fine-tune the parameters to obtain the liquid hydrogen tank pressure prediction model, such as Figure 3 shown.

[0064] Based on the pre-trained model, transfer learning training is carried out. A small sample experimental data set is obtained, and the parameter fine-tuning mechanism in the transfer learning process is specified in the python program to obtain a liquid hydrogen tank pressure prediction model. The small data sample set is selected from five public liquid hydrogen self-pressurization experiments. The pre-trained model is trained using the model-based parameter fine-tuning method in the field of transfer learning, that is: freezing some parameters of the pre-trained model, and only retaining some model parameters to participate in the transfer learning process, such as: fine-tuning the weights and bias of the input layer, the first hidden layer, and the output layer, and freezing the remaining parameters. Using the idea of ​​exhaustive enumeration, the fine-tuning strategies for the input layer, the first hidden layer, and the output layer are further compared, such as: selecting the weights and bias parameters of the input layer, the first hidden layer, and the output layer as activation objects, activating one of the three layers, two of the three layers, or all three. In the process of comparing the seven activation schemes, a five-fold cross-validation method is used to compare the MAE, MSE, and R 2 As an evaluation indicator of the model after transfer learning, the liquid hydrogen storage tank pressure prediction model proposed in the present invention is finally obtained.

[0065] The data results of the self-pressurization test of liquid hydrogen tanks in the publicly published test reports of five liquid hydrogen tanks are shown in Table 1. After unit change and data cleaning, the data is used as a small sample test data set for the transfer learning process, and its features and labels are the same as those of the large sample data set.

[0066] Table 1 Self-pressurization test of five liquid hydrogen storage tanks

[0067]

[0068] 2.3 Pressure prediction

[0069] 1. As Figure 4 As shown, the ratio of the internal surface area of ​​the tank to the total volume of the tank is , Total heat leakage of liquid hydrogen storage tank , filling rate of liquid hydrogen storage tank 、The current gas phase pressure inside the liquid hydrogen storage tank and estimated storage duration As input parameters, a five-dimensional vector is formed. In the Python compilation platform (including: Pycharm, VScode and other software), the trained liquid hydrogen tank pressure prediction model is imported, the input parameters are imported, and the calculation is obtained. The internal pressure of the liquid hydrogen storage tank after time .

[0070]

[0071]

[0072] in, is the input vector, For the trained pressure prediction model, To predict pressure.

[0073] Existing liquid hydrogen tank pressure prediction models usually use thermodynamic models and CFD models. Thermodynamic models need to make assumptions about the two-phase state inside the liquid hydrogen tank and simplify the heat transfer process inside and outside the tank. Since the limited space natural convection phenomenon in the two-phase flow zone is simplified to the limited space large plate natural convection process in thermodynamics, its natural convection heat transfer coefficient is determined by an empirical formula, which leads to a certain error between the model prediction results and the actual results. In addition, the thermodynamic model with fewer assumptions has a higher complexity, and the compilation and calculation process is also more complicated. In contrast, the CFD model can capture the flow and heat transfer details of the gas phase, liquid phase and two-phase region in the tank in more detail. The multiphase flow model VOF is usually combined with the evaporation and condensation Lee phase change model for simulation calculation. However, due to the failure to consider the influence of pressure changes on the heat and mass transfer exchange rate of the phase change process, as well as the adiabatic assumption of the inner wall of the tank, the error gradually increases after a period of calculation. In addition, the CFD model requires a lot of time and resources during the calculation process, and it is difficult to provide accurate pressure prediction values ​​in a short time. The present invention proposes a BP neural network prediction model for liquid hydrogen storage tank pressure based on transfer learning. The model not only avoids modeling of two-phase fluid inside a complex liquid hydrogen storage tank, but also reduces the adverse effect on the generalization performance of the neural network caused by too few data samples.

[0074] The present invention first obtains a large sample data set through a thermodynamic partition model, and uses the initial gas phase pressure in the storage tank, the ratio of the surface area to the total volume of the storage tank, the initial filling rate, the total heat leakage and the storage time as input features. Then, the pre-model is trained in combination with the Bayesian hyperparameter optimization method. Then, the parameters of the pre-model are fine-tuned using a small sample experimental data set, and the exhaustive method is used to obtain the optimal parameter fine-tuning strategy and the corresponding liquid hydrogen storage tank pressure prediction model after transfer learning. Through this model, the current filling situation in the liquid hydrogen storage tank can be quickly and accurately evaluated for safety and economy, providing data support for exhaust and pressure relief decisions during the long-term storage and transportation of liquid hydrogen.

[0075] In view of the problem that the current prediction accuracy of liquid hydrogen storage tank pressure using thermodynamic models and CFD models is insufficient and the too few test data samples limit the generalization ability of neural networks, the present invention provides a liquid hydrogen storage tank pressure prediction model based on transfer learning and a method of use, which trains a high-precision liquid hydrogen storage tank BPNN model through a small amount of data sets. This model not only avoids the modeling of the two-phase fluid inside the complex liquid hydrogen storage tank, but also reduces the adverse effects on the generalization performance of the neural network due to too few data samples. Thereby, it provides strong data support for formulating the pressure relief strategy for the liquid hydrogen storage and transportation process and realizing the economical and safe liquid hydrogen filling and storage process. This method gets rid of the dilemma of low generalization ability of the neural network model when only a small amount of test data sets are used for training, and improves the accuracy of the pressure prediction model.

[0076] Embodiment 2

[0077] Based on the liquid hydrogen storage tank pressure prediction method described in Example 1, this embodiment provides a liquid hydrogen storage tank pressure prediction device, including: a data acquisition module, used to obtain process parameters of the liquid hydrogen storage tank; a pressure prediction module, used to input the acquired process parameters of the liquid hydrogen storage tank into a trained liquid hydrogen storage tank pressure prediction model, and output a predicted value of the tank pressure of the liquid hydrogen storage tank after the expected storage time.

[0078] Embodiment 3

[0079] Based on the liquid hydrogen storage tank pressure prediction method described in Example 1, this embodiment provides a liquid hydrogen storage tank pressure prediction system, including: a storage medium and a processor; the storage medium is used to store instructions; the processor is used to operate according to the instructions to execute the liquid hydrogen storage tank pressure prediction method described in Example 1.

[0080] It should be noted that the entire process of the above-mentioned liquid hydrogen storage tank pressure prediction model and usage method can refer to the aforementioned method embodiment and be implemented in a computer and a computer-based program. To avoid repetition, they will not be described one by one here.

[0081] A person skilled in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant software through a computer program. A person skilled in the art can clearly realize that for the sake of simplicity of the application content and the convenience of description, the above-mentioned examples are only used as supplementary explanations, and in actual applications, the above-mentioned models and usage methods can be split into other models and usage methods with similar functions according to storage media, usage scenarios, and task requirements as needed to realize all or part of the above-mentioned functions.

[0082] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, it should be understood by those skilled in the art that the technical solutions described in the embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention.

Claims

1. A method for predicting the pressure of a liquid hydrogen storage tank, characterized in that: include: Obtain the process parameters of the liquid hydrogen storage tank and perform preprocessing to obtain input parameters; The input parameters are input into the trained liquid hydrogen storage tank pressure prediction model, and the predicted value of the internal pressure of the liquid hydrogen storage tank after the expected storage time is output.

2. The method for predicting the pressure of a liquid hydrogen storage tank according to claim 1, characterized in that: The process parameters of the liquid hydrogen storage tank include: the gas phase pressure inside the liquid hydrogen storage tank at the current moment , Liquid hydrogen temperature , gas hydrogen temperature , Liquid hydrogen volume , External ambient temperature , Total volume of storage tank , Tank internal surface area , Estimated storage time .

3. The method for predicting the pressure of a liquid hydrogen storage tank according to claim 2, characterized in that: The pretreatment includes: calculating the ratio of the internal surface area of ​​the tank to the total volume of the tank according to the process parameters of the liquid hydrogen tank , Total heat leakage of liquid hydrogen storage tank , filling rate of liquid hydrogen storage tank , and the ratio of the internal surface area of ​​the tank to the total volume of the tank , Total heat leakage of liquid hydrogen storage tank , filling rate of liquid hydrogen storage tank 、The current gas phase pressure inside the liquid hydrogen storage tank and estimated storage duration as an input parameter.

4. The method for predicting the pressure of a liquid hydrogen storage tank according to claim 1, characterized in that: The liquid hydrogen storage tank pressure prediction model is constructed based on the BPNN back-propagation neural network model, including: an input layer, multiple hidden layers and an output layer, the input layer includes five input layer nodes, and the output layer includes one output layer node.

5. The method for predicting the pressure of a liquid hydrogen storage tank according to claim 4, characterized in that: The training method of the liquid hydrogen storage tank pressure prediction model comprises: First, a large sample data set is generated using the thermodynamic partition model of a liquid hydrogen storage tank. The number of hidden layers HLN, the number of neurons in each hidden layer HLS, and the initial learning rate ILR are used as hyperparameters. Based on the Bayesian optimization algorithm, the neural network model defined by different hyperparameter combinations is trained and verified to obtain a pre-trained model. The pre-trained model is then trained using a small sample data set based on transfer learning, and the liquid hydrogen storage tank pressure prediction model is obtained by fine-tuning the parameters.

6. The method for predicting the pressure of a liquid hydrogen storage tank according to claim 5, characterized in that: The thermodynamic partition model of liquid hydrogen storage tank adopts the following assumptions: (1) The liquid hydrogen storage tank is divided into three areas: overheating area, saturated area and supercooling area. Each area is a control body with uniform parameters inside the control body. (2) The two-phase saturated zone in the superheated zone, saturated zone and supercooled zone is a massless interface. The physical properties of hydrogen inside the interface are saturated and the pressure is equal to the saturation pressure. (3) The total heat leakage of the liquid hydrogen storage tank is composed of the heat leakage of the wall insulation structure, support structure and pipeline, and the change of heat leakage density with time and space is not considered.

7. The method for predicting the pressure of a liquid hydrogen storage tank according to claim 5, characterized in that: The learning algorithm of the pre-trained model is Adam, and the loss function is MSE function; the cosine annealing algorithm is used to adjust the learning rate of the model under each set of hyperparameter combinations to accelerate the convergence speed of model training; the five-fold cross-validation training method is used to achieve effective estimation of model performance and generalization ability.

8. The method for predicting the pressure of a liquid hydrogen storage tank according to claim 5, characterized in that: The parameter fine-tuning includes: training the pre-trained model using a model-based parameter fine-tuning method, specifically: fine-tuning the weights and biases of the input layer, the first hidden layer and the output layer, and freezing the remaining parameters; adopting an exhaustive idea, further comparing the fine-tuning strategies of the input layer, the first hidden layer and the output layer using a five-fold cross-validation method, and finally obtaining the liquid hydrogen storage tank pressure prediction model.

9. A liquid hydrogen tank pressure prediction device, characterized in that: include: Data acquisition module, used to obtain process parameters of liquid hydrogen storage tanks; The pressure prediction module is used to input the acquired process parameters of the liquid hydrogen storage tank into the trained liquid hydrogen storage tank pressure prediction model, and output the predicted value of the tank pressure after the expected storage time of the liquid hydrogen storage tank.

10. A liquid hydrogen storage tank pressure prediction system, characterized in that: include: Storage media and processors; The storage medium is used to store instructions; The processor is used to operate according to the instruction to execute the liquid hydrogen storage tank pressure prediction method according to any one of claims 1-8.

Images