Dynamic prediction method for cracking and coking amount of supercritical aviation kerosene in pipeline

Abstract

The invention discloses a dynamic prediction method for the cracking and coking amount of supercritical aviation kerosene in a pipeline, and the method comprises the following steps: constructing a cracking and coking data set in combination with a 24-step cracking reaction model and a coking kinetic model of aviation kerosene; a feedback structure is introduced into the topological structure of the neural network, and a cracking coking time sequence prediction model is constructed; training and optimizing the time sequence prediction model based on the cracking and coking data set; and dynamically predicting the cracking and coking amount of the supercritical aviation kerosene in the pipeline based on the optimized time sequence prediction model. According to the method, high accuracy and strong generalization of the prediction result of the dynamic change of the cracking and coking amount are ensured by constructing the cracking and coking data set, optimizing the artificial neural network architecture, adjusting the topological structure of the model and combining the precision verification of the classic time sequence prediction model.

Description

Technical Field

[0001] The invention belongs to the technical field of cracking and coking of aviation kerosene in a hypersonic aircraft, and in particular relates to a method for dynamically predicting the amount of cracking and coking of supercritical aviation kerosene in a pipeline. Background Art

[0002] Hypersonic vehicles refer to new aircraft, new artillery shells, cruise missiles and other winged or wingless aircraft with a flight Mach number greater than 5. They not only reflect the top level of cutting-edge high-tech technology, but are also an important symbol of a new leap in scientific and technological levels. In the propulsion system of hypersonic vehicles, the aviation kerosene in the pipeline not only provides power as fuel, but is also used as a cooling medium to reduce high-temperature components. Under high temperature and high pressure conditions, aviation kerosene reaches a supercritical state, resulting in a cracking reaction to generate coking products such as gaseous hydrocarbons and coke. As the working time continues to accumulate, the flow heat transfer efficiency of the fuel and the service life of the system will be greatly reduced. Therefore, how to accurately predict the dynamic changes in the amount of coking in the pipeline in real time during actual flight is a key technology to ensure the reliability and safety of hypersonic vehicles.

[0003] In the prior art, empirical cracking reaction models or simplified kinetic models are usually used to predict coking. However, these methods have the following defects: 1. Traditional empirical models are difficult to accurately reflect the actual coking dynamic process under complex supercritical conditions, resulting in insufficient prediction accuracy. 2. Existing prediction methods usually estimate the amount of coking under steady state or specific operating conditions, and lack the ability to predict the dynamic changes of coking amount over time. 3. Due to the lack of support from multi-operating condition data, traditional models show poor generalization when faced with different flight conditions or operating condition changes, and are difficult to promote in practical applications.

[0004] In view of the above problems, it is urgent to propose a dynamic prediction method for the amount of coking during cracking of supercritical aviation kerosene in pipelines. Summary of the invention

[0005] In order to solve the above technical problems, the present invention proposes a dynamic prediction method for the amount of coking by cracking supercritical aviation kerosene in a pipeline. Sufficient data samples are constructed through a 24-step cracking reaction model and a coking kinetic model, and a loop layer is introduced into the architecture of a traditional artificial neural network, so that the model can dynamically capture the trend of the amount of coking changing over time. In addition, by combining the optimized model topology structure with the classical time series prediction model, high-precision coking prediction under multiple working conditions is achieved, overcoming the defects of low prediction accuracy, poor dynamic response capability and insufficient model generalization in the prior art.

[0006] To achieve the above object, the present invention provides a method for dynamically predicting the amount of coking by cracking supercritical aviation kerosene in a pipeline, comprising the following steps:

[0007] Combining the 24-step cracking reaction model and coking kinetic model of aviation kerosene, a cracking coking data set was constructed.

[0008] Introducing feedback structure into the topology of neural network, constructing a time series prediction model for cracking coking;

[0009] Training and optimizing the time series prediction model based on the cracking coking data set;

[0010] The amount of coking by cracking of supercritical aviation kerosene in the pipeline is dynamically predicted based on the optimized time series prediction model.

[0011] Optionally, the process of constructing a cracking coking data set by combining the 24-step cracking reaction model and the coking kinetic model of aviation kerosene includes:

[0012] The 24-step cracking reaction model and coking kinetic model of the aviation kerosene are embedded in a computational fluid dynamics solver, and transient simulations are performed on different operating conditions to generate corresponding transient data; time series data of key variables are extracted from the transient data to generate flow, reaction and coking data under multiple transient operating conditions, and an initial cracking and coking data set is constructed; data cleaning and correction are performed on the initial cracking and coking data set; format conversion is performed on the processed initial cracking and coking data set, and the operating conditions and time points corresponding to each data are marked to obtain a final cracking and coking data set.

[0013] Optionally, a feedback structure is introduced into the topological structure of the neural network, and the process of constructing a time series prediction model for cracking coking includes:

[0014] Add a loop layer to the hidden layer of the neural network. The values ​​of the hidden layer and the output layer after adding the loop layer are expressed as:

[0015] S t =f(UgX t +WgS t-1 )

[0016] Y t =g(VgS t )

[0017] Among them, S t represents the value of the hidden layer at time t, f represents the transfer function between the value of the input layer and the loop layer to the hidden layer, U represents the weight matrix between the output layer and the hidden layer, g represents the transfer function between the value of the hidden layer to the output layer, X t represents the value of the input layer at time t, W represents the last value of the hidden layer as the current input weight matrix, S t-1 represents the value of the hidden layer at time t-1; Y trepresents the value of the output layer at time t, and V represents the weight matrix between the hidden layer and the output layer.

[0018] Optionally, before training the time series prediction model based on the cracking coking data set, the method further includes:

[0019] The cracking coke data set is subjected to maximum and minimum normalization processing, and the processed cracking coke data set is randomly divided into a training set, a validation set and a test set.

[0020] Optionally, the process of training the time series prediction model based on the cracking coke data set includes: iteratively processing the time series prediction model based on the training set in the cracking coke data set, and respectively calculating the activation values ​​of the input layer, hidden layer, output layer and loop layer of the time series prediction model. When the activation values ​​of each layer meet the iteration requirements, the training ends.

[0021] Optionally, the process of optimizing the time series prediction model based on the cracking coking data set includes:

[0022] Based on the validation set in the cracking coke data set, the mean square error and relative error of the actual output and expected output of the time series prediction model are calculated. If the mean square error and relative error do not meet the set values, the topological structure of the time series prediction model is optimized; wherein the topological structure to be optimized includes the number of network layers, the number of nodes in each network layer, the activation function, the learning algorithm, the learning rate and the loss function.

[0023] The present invention also provides a dynamic prediction system for the amount of coking by cracking supercritical aviation kerosene in a pipeline, which is used to implement a dynamic prediction method for the amount of coking by cracking supercritical aviation kerosene in a pipeline, comprising: a data acquisition module, an architecture design module, a model optimization module and an accuracy prediction module;

[0024] The data acquisition module is used to combine the 24-step cracking reaction model and the coking kinetic model of aviation kerosene to construct a cracking coking data set;

[0025] The architecture design module is used to introduce a feedback structure into the topological structure of the neural network to construct a time series prediction model for cracking and coking;

[0026] The model optimization module is used to train and optimize the time series prediction model based on the cracking coking data set;

[0027] The accuracy prediction module is used to test the prediction accuracy of the optimized time series prediction model based on the cracking coking data set.

[0028] The present invention also provides an electronic device, comprising: a memory and a processor; the memory is used to store a program; the processor is used to execute the program to implement each step of the dynamic prediction method of the amount of coking by cracking supercritical aviation kerosene in a pipeline.

[0029] The present invention also provides a readable storage medium on which a computer program is stored. When the computer program is executed by a processor, each step of the method for dynamically predicting the amount of coking by cracking supercritical aviation kerosene in a pipeline is implemented.

[0030] Compared with the prior art, the present invention has the following advantages and technical effects:

[0031] The present invention can more accurately capture the complex reaction process under supercritical conditions by constructing a cracking coking data set and a time series prediction model, and significantly improve the accuracy of coking amount prediction; wherein, when constructing the time series prediction model, a loop layer is introduced into the traditional neural network topology structure, so that the time series prediction model can dynamically capture the time evolution law of the coking amount, significantly improve the prediction accuracy of the coking amount at different time points, and solve the defect that the traditional method cannot effectively handle the dynamic change problem; the present invention trains and optimizes the time series prediction model, significantly reduces the prediction error, so that the optimized time series prediction model can still maintain high-precision coking amount prediction under various flight conditions, thereby improving the generalization ability and practicality of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0033] Figure 1 This is a schematic flow chart of a method for predicting the amount of coking formed by cracking supercritical aviation kerosene in a pipeline according to an embodiment of the present invention;

[0034] Figure 2 Schematic diagram of the topological structure of the recurrent neural network according to an embodiment of the present invention. DETAILED DESCRIPTION

[0035] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0036] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0037] like Figure 1 As shown, this embodiment provides a method for dynamically predicting the amount of coking by cracking supercritical aviation kerosene in a pipeline, comprising the following steps:

[0038] Combining the 24-step cracking reaction model and coking kinetic model of aviation kerosene, a cracking coking data set was constructed.

[0039] Introducing feedback structure into the topology of neural network, constructing a time series prediction model for cracking coking;

[0040] Training and optimizing the time series prediction model based on the cracking coking data set;

[0041] The amount of coking by cracking of supercritical aviation kerosene in the pipeline is dynamically predicted based on the optimized time series prediction model.

[0042] Step 1: Combine the 24-step cracking reaction model and coking kinetic model of aviation kerosene to construct a cracking coking data set. The process of embedding the cracking reaction model and coking kinetic model into the computational fluid dynamics simulation software includes:

[0043] The kinetic model of the 24-step molecular reaction of supercritical aviation kerosene contains 1 main cracking reaction and 23 secondary cracking reactions, which is suitable for the case of high cracking degree. The reaction formula of the main cracking reaction is:

[0044] RP-3→0.1086H2+0.4773CH4+0.5586C2H4+0.39C2H6+0.41C3H6

[0045] +0.2001C3H8+0.2246C4H8+0.0353C4H 10 +0.031C4H6

[0046] +0.7201C 5+ +0.27CC 5+ +0.0222C n H 2n-6

[0047] The cleavage reaction rate constant k c The calculation formula uses the form of Arrenhius's law:

[0048]

[0049] In the formula, the pre-exponential factor A c =2.869×10 14 / s, activation energy E a =217.9 kJ / mol, constant R =8.314 J / (mol·K).

[0050] Table 1 lists the remaining 23 secondary cracking reactions in the fuel cracking model and the corresponding pre-exponential factors and activation energies in the Arrhenius formula.

[0051] Table 1

[0052]

[0053]

[0054] The coking kinetic model uses propylene and aromatics in the cracking products of supercritical aviation kerosene as coking matrix, and the expression of the wall coking model is as follows:

[0055]

[0056] J c is the wall coking rate, that is, the wall coking mass per unit time and unit area; the four variables C p1 , C p2 , T w,i and t are the molar concentration of propylene, the molar concentration of aromatic hydrocarbons, the temperature of the inner wall of the pipeline and the duration of the coking process, respectively; the activation energy E a1 =232.3 kJ / mol, E a2 =80.5 kJ / mol, pre-exponential factor A c1 =1.16×10 15 μg·cm -2 ·min -1 / (mol·L -1 ) n , A c2 =2.21×10 5 μg·cm -2 ·min -1 / (mol·L -1 ) n , reaction order n1 = 1.55, n2 = 0.14; considering that the metal catalytic elements on the pipeline surface are covered as the coking increases during the catalytic coking process, the catalytic coking rate will decay with time, and the decay factor γ = 0.14.

[0057] According to the constructed cracking reaction model and coking kinetic model, the rate expressions of cracking and coking reactions are defined by embedding the user-defined function (UDF) into the computational fluid dynamics (CFD) solver; the thermophysical properties and turbulence models of reactants, products and intermediates are set; the boundary conditions such as inlet, outlet and wall are set; the transient solver, time step and convergence criterion are set.

[0058] Step 2: Obtain transient data sets under different working conditions and construct cracking coking data sets:

[0059] This embodiment takes a slender horizontal circular pipe as an example, with a pipe length of L = 500mm and a pipe diameter of d = 2mm. A 100mm insulation section is set at the inlet and outlet to ensure full development of the flow, thereby avoiding the inlet and outlet effect. The middle 300mm is the heating section, the system pressure P = 5MPa, and the inlet temperature T in =473K, mass flow rate is m=1g / s, and two boundary conditions are applied, namely, different wall temperatures and wall heat flux. Among them, the wall temperatures are 900K, 950K, 1000K, 1050K and 1100K, and the wall heat flux under the wall heat flux condition is 0.8WM / m 2 、0.9WM / m 2 , 1.0WM / m 2 , 1.1WM / m 2 and 1.2WM / m 2 .

[0060] First, use CFD software to perform transient simulations on different working conditions, and use the batch processing function of the automation tool to generate a large amount of transient data. Secondly, use post-processing tools to extract time series data of key variables and generate flow, reaction and coking data under multiple transient conditions. Next, check the outliers and missing values ​​in the data set, and perform necessary data cleaning and correction. Finally, format the data into a data set that is easy to analyze, and mark the working conditions and time points corresponding to each data point. In this embodiment, data from 1 to 15 minutes in the pipeline are collected for model learning and training, and data from 16 to 20 minutes are used to verify the accuracy of the model. The verification indicator is the root mean square error (RMSE) to measure the predicted value y pre,i and the actual value y act,i The deviation is as follows:

[0061]

[0062] Step 3: Perform maximum and minimum normalization processing on the cracking coke data set:

[0063] Normalization is the process of mapping data feature values ​​to a relatively small and consistent range, which is convenient for subsequent analysis or modeling. It can improve the convergence speed during model training, avoid dimensional differences in features, and prevent problems such as unstable or overflow numerical calculations.

[0064] This example uses maximum and minimum value normalization to process the cracking coke data set constructed in step 2, and the data is scaled to the range of -1 to 1, that is:

[0065]

[0066] X* is the normalized data, which converts the original data X to a value between 0 and 1, where x m is the maximum value of the sample data, x n is the minimum value of the sample data.

[0067] Step 4: Dataset division:

[0068] This embodiment randomly divides the data set processed by the maximum and minimum normalization in step three into a training set, a validation set, and a test set, accounting for 70%, 20%, and 10% of the data set respectively. Among them, the training set is used to train the model, that is, to adjust the parameters of the model so that the model can learn rules from the data; the validation set is used to tune the model parameters and evaluate the performance of the model on new data, helping to determine whether the model is overfitting, so as to select the best model configuration; the test set is data that has never been used in the model training and verification process, so it provides an independent data set to measure the generalization ability of the model, and is used for the final evaluation of the model after the model training and tuning are completed.

[0069] Step 5: Design a timing prediction model:

[0070] This embodiment introduces a feedback structure into the topology of a traditional artificial neural network to design a time series prediction model for cracking coke, that is, a recurrent layer is added to the input layer, hidden layer, and output layer of the model, so that the model can model sequences of any length and use previous information to affect subsequent calculations. Figure 2 As shown, X and Y represent the values ​​of the input layer and output layer respectively, S t represents the value of the hidden layer at time t, U and V represent the weight matrices from the output layer to the hidden layer and from the hidden layer to the output layer respectively, and W represents the last value of the hidden layer as the current input weight matrix. Therefore, the values ​​of the hidden layer and the output layer in the topology can be expressed as:

[0071] S t =f(UgX t +WgS t-1 ) (5)

[0072] Y t =g(VgS t ) (6)

[0073] Where f represents the transfer function from the input layer and the circulation layer to the hidden layer, and g represents the transfer function from the hidden layer to the output layer. In this method, the input layer contains working condition information, geometric information and time variables, and the output layer is the cracking coking amount.

[0074] Step 6: Initialization of model parameters:

[0075] Weights are parameters for connecting nodes, which are used to adjust the strength of input signals. Each input connected to a node will have a corresponding weight. Threshold is an additional parameter of the node, which is used to adjust the input of the activation function. It allows the network to have non-zero output without any input signal. The initialization of model weights and thresholds is crucial to the training process. Proper initialization can help accelerate convergence, improve model performance, and avoid gradient vanishing or exploding problems. This method uses He random initialization for the weights and thresholds of the model in step 4, which is suitable for the ReLU activation function. The initialization mean is 0, and the variance is set according to the number of nodes in the previous layer of the network. Its law follows the normal distribution.

[0076] Step 7: Model training and validation:

[0077] Use the training set in step 4 to iterate the model in step 6 and calculate the activation values ​​of the model input layer, hidden layer, output layer and loop layer at time T. The activation value of the input layer is usually the input data itself. Assume that the input at time T is x T ,but:

[0078]

[0079] The calculation of the activation value of the hidden layer involves the following steps: Assume that the activation value of the lth layer is a T (l-1) , the weight matrix is ​​W (l) , the bias vector is b (l) , the activation function is φ, then the activation value a of the lth layer T (l) Calculated as:

[0080]

[0081]

[0082] The activation value calculation of the loop layer involves the current input x T and the hidden state h at the previous time step T-1 , assuming that the weight matrix of the hidden layer is W h , for the current hidden state, W x , used for input data, biased by b h , the activation function is φ, then the hidden state h of the recurrent layer T The calculation is as follows:

[0083] z T =W h h T-1 +W x x T +b h (10)

[0084] h T =φ(z T ) (11)

[0085] The activation value of the output layer of the model is calculated as follows: Assume that the weight matrix of the output layer is W out , with a bias of b out , the activation function is φ output , then the activation value y of the output layer T Calculated as:

[0086]

[0087] The activation values ​​between nodes in each layer of the model are used to monitor the training loss, that is, the loss on the training set tends to be stable and changes very little over multiple training cycles, that is, the relative change rate is less than 10 -4 , which usually indicates that the model has converged.

[0088] Step 8: Model optimization:

[0089] Use the validation set in step 4 to calculate the mean square error and relative error between the actual output and the expected output of the model. If the error does not meet the set value, set the mean square error to less than 5 and the relative error to less than 5 μg / cm 2 , the topological structure of the model is optimized, including the number of network layers, the number of nodes in each layer, the activation function, the learning algorithm, the learning rate and the loss function. In this method, the number of layers of the optimized network is 1, the number of nodes is 100, the ReLU activation function is used between the structural units of the model, the Levenberg-Marquardt learning algorithm is used, the Adam optimizer is used to automatically adjust the learning rate, and the loss function is the mean square error loss.

[0090] Table 2 shows the accuracy of the model in predicting different wall temperatures and wall heat flux conditions. It can be found that under ten different wall temperatures and wall heat flux conditions, the RMSE of the T-ANN model predictions from 16 to 20 minutes is less than 5, and the relative error is less than 4 μg / cm 2 , meeting the prediction accuracy requirements.

[0091] Table 2

[0092]

[0093] Step 9: Comparison of the prediction accuracy of cracking coking amount by different models:

[0094] Compare and analyze the accuracy of two commonly used time series prediction models and the T-ANN model established in step 8. ARMA (Autoregressive Moving Average), also known as the autoregressive moving average prediction model, is used to describe the autocorrelation and sliding average relationship of time series data. It is a statistical model that combines the autoregressive model and the sliding average model. It represents the predicted value at time t+1, and its mathematical formula is as follows:

[0095]

[0096] Among them, p and q are orders, γ i is the autoregressive coefficient, θ i is the moving average coefficient, ε t For error.

[0097] ES (Exponential Smoothing) time series forecasting model is also called exponential smoothing model. It is based on the weighted average of time series data. It predicts future observations by weighting historical observations exponentially. The basic principle is to assume that the prediction of future observations is the weighted average of past observations, where the weight decreases exponentially. The mathematical formula is as follows:

[0098]

[0099] In the formula,

[0100]

[0101]

[0102] in, and They represent the first, second and third exponential smoothing values ​​at time t respectively. The mathematical formulas are as follows:

[0103]

[0104] Among them, α is the smoothing coefficient (0<α<1), x t Represents the actual value at time t.

[0105] Table 3 lists the relative errors of the four models with respect to the wall coking amount at five moments. It can be seen from the table that the T-ANN model constructed by this method has extremely high accuracy in predicting the wall coking amount. The relative errors predicted within 16 to 20 minutes are smaller than those of the ARMA model and the ES model, and the relative errors at each moment are all within 1%.

[0106] Table 3

[0107]

[0108] Based on the technical solution of this embodiment, the accurate prediction of the dynamic change of supercritical aviation kerosene cracking coking amount over time in the pipeline of hypersonic aircraft is realized. Through rich data samples and accurate reaction models, this embodiment can more accurately capture the complex reaction process under supercritical conditions, significantly improving the accuracy of coking amount prediction; after introducing the circulation layer in the traditional neural network topology, the model can dynamically capture the time evolution law of coking amount, significantly improve the prediction accuracy of coking amount at different time points, and solve the defect of traditional methods that cannot effectively handle dynamic change problems; through model optimization, the prediction error is significantly reduced, so that the model can still maintain high-precision coking amount prediction under various flight conditions, thereby improving the generalization ability and practicality of the model; after accuracy verification, the T-ANN model proposed in this embodiment maintains stable prediction accuracy over a long period of time, especially under complex dynamic conditions, showing strong generalization ability and adaptability.

[0109] On the other hand, based on the same inventive concept as the above embodiment, this embodiment also provides a dynamic prediction system for the amount of coking by cracking supercritical aviation kerosene in a pipeline. The use method of the system and the dynamic prediction method for the amount of coking by cracking supercritical aviation kerosene in a pipeline provided by the above embodiment can be mutually referenced in terms of effect. The dynamic prediction system includes: a data acquisition module, an architecture design module, a model optimization module and an accuracy prediction module;

[0110] The data acquisition module is used to combine the 24-step cracking reaction model and coking kinetic model of aviation kerosene to construct a cracking coking data set;

[0111] The architecture design module is used to introduce feedback structure into the topology of the neural network and build a time series prediction model for cracking coking;

[0112] The model optimization module is used to train and optimize the time series prediction model based on the cracking coking data set;

[0113] The accuracy prediction module is used to test the prediction accuracy of the optimized time series prediction model based on the cracking coking data set.

[0114] This embodiment also provides an electronic device, including: a memory and a processor; the memory is used to store a program; the processor is used to execute the program to implement each step of the dynamic prediction method of the amount of coking of supercritical aviation kerosene cracking in a pipeline.

[0115] This embodiment also provides a readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, each step of the method for dynamically predicting the amount of coking formed by cracking of supercritical aviation kerosene in a pipeline is implemented.

[0116] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for dynamically predicting the amount of coking during cracking of supercritical aviation kerosene in a pipeline, characterized in that: The following steps are involved: Combining the 24-step cracking reaction model and coking kinetic model of aviation kerosene, a cracking coking data set was constructed. Introducing feedback structure into the topology of neural network, constructing a time series prediction model for cracking coking; Training and optimizing the time series prediction model based on the cracking coking data set; The amount of coking by cracking of supercritical aviation kerosene in the pipeline is dynamically predicted based on the optimized time series prediction model.

2. The method for dynamically predicting the amount of coking by cracking supercritical aviation kerosene in a pipeline according to claim 1, characterized in that: Combining the 24-step cracking reaction model and coking kinetic model of aviation kerosene, the process of constructing the cracking coking data set includes: The 24-step cracking reaction model and coking kinetic model of the aviation kerosene are embedded in a computational fluid dynamics solver, and transient simulations are performed on different operating conditions to generate corresponding transient data; time series data of key variables are extracted from the transient data to generate flow, reaction and coking data under multiple transient operating conditions, and an initial cracking and coking data set is constructed; data cleaning and correction are performed on the initial cracking and coking data set; format conversion is performed on the processed initial cracking and coking data set, and the operating conditions and time points corresponding to each data are marked to obtain a final cracking and coking data set.

3. The method for dynamically predicting the amount of coking by cracking supercritical aviation kerosene in a pipeline according to claim 1, characterized in that: The process of introducing feedback structure into the topological structure of neural network and constructing the time series prediction model of cracking coking includes: Add a loop layer to the hidden layer of the neural network. The values ​​of the hidden layer and the output layer after adding the loop layer are expressed as: S t =f(UgX t +WgS t-1 ) Y t =g(VgS t ) Among them, S t represents the value of the hidden layer at time t, f represents the transfer function between the value of the input layer and the loop layer to the hidden layer, U represents the weight matrix between the output layer and the hidden layer, g represents the transfer function between the value of the hidden layer to the output layer, X t represents the value of the input layer at time t, W represents the last value of the hidden layer as the current input weight matrix, S t-1 represents the value of the hidden layer at time t-1; Y t represents the value of the output layer at time t, and V represents the weight matrix between the hidden layer and the output layer.

4. The method for dynamically predicting the amount of coking by cracking supercritical aviation kerosene in a pipeline according to claim 1, characterized in that: Before training the time series prediction model based on the cracking coking data set, the following step is further included: The cracking coke data set is subjected to maximum and minimum normalization processing, and the processed cracking coke data set is randomly divided into a training set, a validation set and a test set.

5. The method for dynamically predicting the amount of coking by cracking supercritical aviation kerosene in a pipeline according to claim 4, characterized in that: The process of training the time series prediction model based on the cracking coke data set includes: iteratively processing the time series prediction model based on the training set in the cracking coke data set, and calculating the activation values ​​of the input layer, hidden layer, output layer and loop layer of the time series prediction model respectively. When the activation values ​​of each layer meet the iteration requirements, the training ends.

6. The method for dynamically predicting the amount of coking by cracking supercritical aviation kerosene in a pipeline according to claim 5, characterized in that: The process of optimizing the time series prediction model based on the cracking coking data set includes: Based on the validation set in the cracking coke data set, the mean square error and relative error of the actual output and expected output of the time series prediction model are calculated. If the mean square error and relative error do not meet the set values, the topological structure of the time series prediction model is optimized; wherein the topological structure to be optimized includes the number of network layers, the number of nodes in each network layer, the activation function, the learning algorithm, the learning rate and the loss function.

7. A dynamic prediction system for the amount of coking produced by cracking supercritical aviation kerosene in a pipeline, characterized in that: A method for dynamically predicting the amount of coking formed by cracking supercritical aviation kerosene in a pipeline according to any one of claims 1 to 6, comprising: a data acquisition module, an architecture design module, a model optimization module and an accuracy prediction module; The data acquisition module is used to combine the 24-step cracking reaction model and the coking kinetic model of aviation kerosene to construct a cracking coking data set; The architecture design module is used to introduce a feedback structure into the topological structure of the neural network to construct a time series prediction model for cracking and coking; The model optimization module is used to train and optimize the time series prediction model based on the cracking coking data set; The accuracy prediction module is used to test the prediction accuracy of the optimized time series prediction model based on the cracking coking data set.

8. An electronic device, characterized in that: include: Memory and processor; The memory is used to store programs; The processor is used to execute the program to implement each step of the method for dynamically predicting the amount of coking by cracking of supercritical aviation kerosene in a pipeline as described in any one of claims 1 to 6.

9. A readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, each step of the method for dynamically predicting the amount of coking by cracking supercritical aviation kerosene in a pipeline as described in any one of claims 1 to 6 is implemented.

Images