A dynamic prediction method for coking amount of supercritical aviation kerosene cracking in pipelines

Abstract

The application discloses a kind of dynamic prediction methods of supercritical aviation kerosene cracking coking quantity in pipeline, comprising the following steps: combining the 24-step cracking reaction model and coking kinetics model of aviation kerosene, construct cracking coking dataset;Feedback structure is introduced in the topological structure of neural network, construct the time series prediction model of cracking coking;The time series prediction model is trained and optimized based on the cracking coking dataset;Based on the time series prediction model after optimization, the dynamic prediction of supercritical aviation kerosene cracking coking quantity in pipeline is carried out.The application constructs cracking coking dataset, optimizes artificial neural network architecture, adjusts the topological structure of model, and combines the accuracy verification of classic time series prediction model, guarantees the high accuracy and strong generalization of the dynamic change prediction result of cracking coking quantity.

Description

Technical Field

[0001] This invention belongs to the technical field of aviation kerosene cracking and coking in hypersonic vehicles, and particularly relates to a dynamic prediction method for the amount of supercritical aviation kerosene cracking and coking in pipelines. Background Technology

[0002] Hypersonic vehicles refer to new types of aircraft, artillery shells, cruise missiles, and other winged or wingless aircraft that fly at Mach numbers greater than 5. They not only represent the pinnacle of cutting-edge high-tech technology but also signify a significant leap forward in scientific and technological advancement. In the propulsion system of hypersonic vehicles, aviation kerosene in the pipelines serves not only as fuel to provide power but also as a cooling medium to reduce the temperature of components. Under high temperature and pressure conditions, aviation kerosene reaches a supercritical state, undergoing a cracking reaction that produces gaseous hydrocarbons, coke, and other coking products. With continuous accumulation over time, this significantly reduces the fuel's flow heat exchange efficiency and the system's lifespan. Therefore, accurately and in real-time predicting the dynamic changes in the amount of coking within the pipeline during actual flight is a key technology for ensuring the reliability and safety of hypersonic vehicles.

[0003] In existing technologies, empirical cracking reaction models or simplified kinetic models are commonly used for coking prediction. However, these methods have the following drawbacks: 1. Traditional empirical models struggle to accurately reflect the actual coking dynamics under complex supercritical conditions, resulting in insufficient prediction accuracy. 2. Existing prediction methods typically estimate coking amounts under steady-state or specific operating conditions, lacking the ability to predict the dynamic changes in coking amounts over time. 3. Due to the lack of multi-condition data support, traditional models exhibit poor generalization ability when faced with different flight conditions or varying operating conditions, making them difficult to promote in practical applications.

[0004] To address the aforementioned issues, there is an urgent need to propose a dynamic prediction method for the amount of coking in supercritical aviation kerosene cracking within pipelines. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention proposes a dynamic prediction method for coking amount in supercritical aviation kerosene pyrolysis within pipelines. Sufficient data samples are constructed using a 24-step pyrolysis reaction model and a coking kinetics model. A recurrent layer is introduced into the traditional artificial neural network architecture, enabling the model to dynamically capture the trend of coking amount changes over time. Furthermore, by combining the optimized model topology with a classic time-series prediction model, high-precision coking prediction under multiple operating conditions is achieved, overcoming the shortcomings of low prediction accuracy, poor dynamic response, and insufficient model generalization in existing technologies.

[0006] To achieve the above objectives, this invention provides a dynamic prediction method for the amount of coking in supercritical aviation kerosene cracking within a pipeline, comprising the following steps:

[0007] A pyrolysis and coking dataset was constructed by combining a 24-step pyrolysis reaction model and a coking kinetics model for aviation kerosene.

[0008] A feedback structure is introduced into the topology of the neural network to construct a time-series prediction model for decomposed coke.

[0009] The time-series prediction model is trained and optimized based on the pyrolysis and coking dataset.

[0010] Dynamic prediction of coking amount in supercritical aviation kerosene cracking within pipelines is performed based on an optimized time-series prediction model.

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

[0012] The 24-step cracking reaction model and coking kinetic model of aviation kerosene are embedded into the solver of computational fluid dynamics to perform transient simulations under different operating conditions, generating 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, constructing an initial cracking and coking dataset. The initial cracking and coking dataset is then cleaned and corrected. The processed initial cracking and coking dataset is then formatted and labeled with the operating conditions and time points corresponding to each data point to obtain the final cracking and coking dataset.

[0013] Optionally, the process of introducing a feedback structure into the topology of the neural network to construct a time-series prediction model for decomposed coke includes:

[0014] Adding a recurrent layer to the hidden layer of a neural network, the values ​​of the hidden layer and the output layer after adding the recurrent layer are represented as follows:

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

[0016] Y t =g(VgS) t )

[0017] Among them, S t Let X represent the value of the hidden layer at time t, f represent the transfer function from the input layer and recurrent layer to the hidden layer, U represent the weight matrix from the output layer to the hidden layer, g represent the transfer function from the hidden layer to the output layer, and X represent the value of the hidden layer at time t. t Let S represent the value of the input layer at time t, W represent the previous value of the hidden layer used as the current input weight matrix, and S represent the value of the input layer at time t. t-1 Y represents the value of the hidden layer at time t-1; tLet t represent the value of the output layer at time t, and V represent the weight matrix between the hidden layer and the output layer.

[0018] Optionally, before training the time-series prediction model based on the ruptured coke dataset, the method further includes:

[0019] The pyrrhic coke dataset is subjected to max-min normalization, and the processed pyrrhic coke dataset is randomly divided into training set, validation set and test set.

[0020] Optionally, the process of training the time series prediction model based on the ruptured coke dataset includes: iteratively processing the time series prediction model based on the training set in the ruptured coke dataset, calculating the activation values ​​of the input layer, hidden layer, output layer and recurrent layer of the time series prediction model respectively, and ending the training when the activation values ​​of each layer meet the iteration requirements.

[0021] Optionally, the process of optimizing the time-series prediction model based on the rupture and coking dataset includes:

[0022] Based on the validation set in the fission coke dataset, 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 topology of the time series prediction model is optimized. The topology to be optimized includes 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.

[0023] This invention also provides a dynamic prediction system for the amount of coking in supercritical aviation kerosene cracking in pipelines, used to implement a dynamic prediction method for the amount of coking in supercritical aviation kerosene cracking in pipelines, including: 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 construct a pyrolysis and coking dataset by combining a 24-step pyrolysis reaction model and a coking kinetics model for aviation kerosene.

[0025] The architecture design module is used to introduce a feedback structure into the topology of the neural network to build a time-series prediction model for decomposition and coagulation.

[0026] The model optimization module is used to train and optimize the time-series prediction model based on the pyrolysis and coking dataset;

[0027] The accuracy prediction module is used to test the prediction accuracy of the optimized time-series prediction model based on the pyrolysis and coking dataset.

[0028] The present invention also provides an electronic device, comprising: a memory and a processor; the memory for storing a program; the processor for executing the program to realize the various steps of the dynamic prediction method for coking amount of supercritical aviation kerosene cracking in pipelines.

[0029] The present invention also provides a readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the various steps of a method for dynamically predicting the amount of coking in supercritical aviation kerosene cracking within a pipeline.

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

[0031] This invention, by constructing a fission coking dataset and a time-series prediction model, can more accurately capture complex reaction processes under supercritical conditions, significantly improving the accuracy of coking quantity prediction. Specifically, in constructing the time-series prediction model, a recurrent layer is introduced into the traditional neural network topology, enabling the model to dynamically capture the temporal evolution of coking quantity, significantly improving the prediction accuracy for coking quantity at different time points and overcoming the shortcomings of traditional methods in effectively handling dynamic changes. Furthermore, this invention trains and optimizes the time-series prediction model, significantly reducing prediction errors. The optimized model maintains high-precision coking quantity prediction under various flight conditions, thereby enhancing the model's generalization ability and practicality. Attached Figure Description

[0032] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0033] Figure 1 This is a schematic diagram of the method for predicting coking amount in supercritical aviation kerosene cracking in pipelines according to an embodiment of the present invention.

[0034] Figure 2 This is a schematic diagram of the topology of a recurrent neural network according to an embodiment of the present invention. Detailed Implementation

[0035] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

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

[0037] like Figure 1 As shown, this embodiment provides a dynamic prediction method for the amount of coking in supercritical aviation kerosene cracking within a pipeline, including the following steps:

[0038] A pyrolysis and coking dataset was constructed by combining a 24-step pyrolysis reaction model and a coking kinetics model for aviation kerosene.

[0039] A feedback structure is introduced into the topology of the neural network to construct a time-series prediction model for decomposed coke.

[0040] The time-series prediction model is trained and optimized based on the pyrolysis and coking dataset.

[0041] Dynamic prediction of coking amount in supercritical aviation kerosene cracking within pipelines is performed based on an optimized time-series prediction model.

[0042] Step 1: Construct a pyrolysis and coking dataset by combining the 24-step pyrolysis reaction model and coking kinetics model for aviation kerosene. The process of embedding the pyrolysis reaction model and coking kinetics model into computational fluid dynamics simulation software includes:

[0043] The kinetic model of the 24-step molecular reaction of supercritical aviation kerosene includes one main cracking reaction and 23 secondary cracking reactions, and is applicable to high cracking degrees. The reaction formula of the main cracking reaction is as follows:

[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] Pyrolysis reaction rate constant k c The calculation formula adopts 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 =217.9kJ / mol, constant R =8.314J / (mol·K).

[0050] Table 1 lists the remaining 23 secondary pyrolysis reactions in the fuel pyrolysis model, along with the pre-exponential factor and activation energy in the corresponding Arrhenius equation.

[0051] Table 1

[0052]

[0053]

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

[0055]

[0056] J c The wall coking rate is defined as the mass of wall coking per unit time and per unit area; four variables C p1 C p2 T w,i t and t represent the molar concentration of propylene (the coking precursor), the molar concentration of aromatics, the temperature of the inner wall of the pipeline, and the time of the coking process, respectively; activation energy E a1 =232.3kJ / 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 The reaction order n1 = 1.55 and n2 = 0.14. Considering that the metal catalytic elements on the pipe surface are covered as coking increases during the catalytic coking process, the catalytic coking rate will decrease over time, with an attenuation factor γ = 0.14.

[0057] Based on the constructed cracking reaction model and coking kinetic model, a user-defined function (UDF) is embedded into the solver of computational fluid dynamics (CFD) to define the rate expressions for cracking and coking reactions; set the thermophysical properties and turbulence models of reactants, products, and intermediates; set boundary conditions such as inlet, outlet, and wall; and set the transient solver, time step, and convergence criteria.

[0058] Step 2: Obtain transient datasets under different operating conditions and construct a pyrolysis and coking dataset:

[0059] This embodiment uses a slender horizontal circular pipe as an example, with a pipe length L = 500 mm and a pipe diameter d = 2 mm. Insulation sections of 100 mm are provided at both the inlet and outlet to ensure sufficient flow development, thereby avoiding inlet / outlet effects. The middle 300 mm section is the heating section. The system pressure P = 5 MPa and the inlet temperature T... in =473K, mass flow rate m = 1g / s, two boundary conditions were applied: different wall temperatures and wall heat flux. The wall temperatures were 900K, 950K, 1000K, 1050K, and 1100K, and the wall heat flux under the wall heat flux condition was 0.8W / m. 2 0.9WM / m 2 1.0WM / m 2 1.1 WM / m 2 and 1.2WM / m 2 .

[0060] First, CFD software was used to perform transient simulations under different operating conditions, generating a large amount of transient data using the batch processing function of automated tools. Second, post-processing tools were used to extract time-series data of key variables, generating flow, reaction, and coking data under multiple transient operating conditions. Next, outliers and missing values ​​in the dataset were checked, and necessary data cleaning and corrections were performed. Finally, the data was formatted into an easily analyzable dataset, and the operating conditions and time points corresponding to each data point were labeled. In this embodiment, data from 1 to 15 minutes within the pipeline were collected for model learning and training, and data from 16 to 20 minutes were used for model accuracy validation. The validation metric was the root mean square error (RMSE) measuring the predicted value y. pre,i and actual value y act,i The deviation is calculated using the following formula:

[0061]

[0062] Step 3: Perform max-min normalization on the pyrolysis and coking dataset:

[0063] Normalization maps data feature values ​​to a relatively small and consistent range, which facilitates subsequent analysis or modeling. It can improve the convergence speed during model training, avoid differences in the dimensions of features, and prevent problems such as instability or overflow in numerical calculations.

[0064] This embodiment uses maximum-minimum normalization to process the pyrolysis and coking dataset constructed in step two, scaling the data to the range of -1 to 1, i.e.:

[0065]

[0066] X* represents the normalized data, which transforms the original data X to values ​​within the range of 0 to 1, where x... m x is the maximum value of the sample data. n This represents the minimum value of the sample data.

[0067] Step 4: Dataset partitioning:

[0068] In this embodiment, the dataset processed by max-min normalization in step three is randomly divided into a training set, a validation set, and a test set, accounting for 70%, 20%, and 10% of the dataset, respectively. The training set is used to train the model, i.e., to adjust the model's parameters so that the model can learn patterns from the data; the validation set is used to fine-tune the model parameters and evaluate the model's performance on new data, helping to determine if the model is overfitting and thus selecting the optimal model configuration; the test set is data that has never been used during model training and validation, therefore providing an independent dataset to measure the model's generalization ability, and is used for the final evaluation of the model after training and fine-tuning.

[0069] Step 5: Design the time series prediction model:

[0070] This embodiment introduces a feedback structure into the traditional topology of artificial neural networks to design a time-series prediction model for splitting and coagulation. Specifically, recurrent layers are added to the input, hidden, and output layers of the model, enabling it to model sequences of arbitrary length and utilize previous information to influence subsequent calculations. The topology of the recurrent neural network is as follows: Figure 2 As shown, X and Y represent the values ​​of the input layer and output layer, respectively, and S... t Let represent the value of the hidden layer at time t, U and V represent the weight matrices between the output layer and the hidden layer, and between the hidden layer and the output layer, respectively, and W represent the weight matrix where the previous value of the hidden layer is used as the current input weight matrix. Therefore, the values ​​of the hidden layer and the output layer in the topology can be represented as:

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

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

[0073] In the formula, f represents the transfer function from the input layer and the recurrent 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 operating condition information, geometric information, and time variables, and the output layer is the amount of cracking and coking.

[0074] Step Six: Initialize Model Parameters:

[0075] Weights are parameters that connect nodes and are used to adjust the strength of the input signal. Each input connected to a node has a corresponding weight. Thresholds are additional parameters for nodes, used to adjust the input to the activation function. They allow the network to have a non-zero output even without any input signal. The initialization of model weights and thresholds is crucial for the training process. Proper initialization can help accelerate convergence, improve model performance, and avoid gradient vanishing or exploding problems. In this method, the model weights and thresholds in step four are randomly initialized using He, suitable for the ReLU activation function. The initial mean is 0, and the variance is set according to the number of nodes in the previous layer of the network, following a normal distribution.

[0076] Step 7: Model Training and Validation

[0077] The training set from step four is used to iterate the model from step six, calculating the activation values ​​of the input layer, hidden layer, output layer, and recurrent layer at time T. The activation values ​​of the input layer are typically the input data itself; assuming the input at time T is x... T ,but:

[0078]

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

[0080]

[0081]

[0082] The activation value calculation of the recurrent layer involves the current input x T The hidden state h of the previous time step T-1 Assume the weight matrix of the hidden layer is W. h Used for the current hidden state, W x , used for input data, biased by b h If the activation function is φ, then the hidden state h of the recurrent layer is... 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 values ​​of the model's output layer are calculated as follows: Assume the weight matrix of the output layer is W. out The bias is b out The activation function is φ output Then the activation value y of the output layer T The calculation is as follows:

[0086]

[0087] The training loss is monitored by the activation values ​​between nodes in each layer of the model. This means that the loss on the training set tends to stabilize and changes very little over multiple training epochs, i.e., the relative rate of change is less than 10. -4 This usually indicates that the model has converged.

[0088] Step 8: Model Optimization

[0089] Using the validation set from step four, calculate the mean square error and relative error of the model's actual and expected outputs. If the errors do not meet the set values, set the mean square error to be less than 5 and the relative error to be less than 5 μg / cm³. 2 The optimization then focuses on the model's topology, specifically including the number of network layers, the number of nodes per layer, the activation function, the learning algorithm, the learning rate, and the loss function. In this method, the optimized network has 1 layer and 100 nodes. The ReLU activation function is used between the model's structural units, the Levenberg-Marquardt learning algorithm is employed, the Adam optimizer is used to automatically adjust the learning rate, and the loss function is the mean squared error loss.

[0090] Table 2 shows the accuracy of the model in predicting different wall temperatures and wall heat flows. It can be observed that under ten different wall temperature and heat flow conditions, the RMSE of the T-ANN model at 16–20 min is less than 5, and the relative error is less than 4 μg / cm³. 2 This meets the requirements for accurate prediction.

[0091] Table 2

[0092]

[0093] Step Nine: Comparison of prediction accuracy for cracking and coking volume using different models:

[0094] This paper compares and analyzes the accuracy of two commonly used time series forecasting models with the T-ANN model established in step eight. ARMA (Autoregressive Moving Average), also known as an autoregressive moving average forecasting model, is used to describe the autocorrelation and moving average relationships in time series data. It is a statistical model that combines autoregressive and moving average models to... Let represent the predicted value at time t+1, and its mathematical formula is as follows:

[0095]

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

[0097] Exponential Smoothing (ES) time series forecasting models, also known as exponential smoothing models, are based on a weighted average of time series data. They predict future observations by exponentially weighting historical observations. The basic principle is that the prediction of future observations is a weighted average of past observations, where the weights decrease exponentially. The mathematical formula is as follows:

[0098]

[0099] In the formula,

[0100]

[0101]

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

[0103]

[0104] Where α is the smoothing coefficient (0 < α < 1), x t This represents the actual value at time t.

[0105] Table 3 lists the relative errors of the four models with the wall coking amount at five time points. As can be seen from the table, 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 all smaller than those of the ARMA model and the ES model, and the relative errors at each time point are all within 1%.

[0106] Table 3

[0107]

[0108] Based on the technical solution of this embodiment, accurate prediction of the dynamic change of coking amount in supercritical aviation kerosene cracking within hypersonic vehicle pipelines over time is achieved. This embodiment, through rich data samples and a precise reaction model, can more accurately capture the complex reaction process under supercritical conditions, significantly improving the accuracy of coking amount prediction. By introducing a recurrent layer into the traditional neural network topology, the model can dynamically capture the temporal evolution of coking amount, significantly improving the prediction accuracy for coking amount at different time points and overcoming the shortcomings of traditional methods in effectively handling dynamic changes. Through model optimization, prediction errors are significantly reduced, enabling the model to maintain high-precision coking amount prediction under various flight conditions, thereby improving the model's generalization ability and practicality. Accuracy verification shows that the T-ANN model proposed in this embodiment maintains stable prediction accuracy over long periods, especially demonstrating strong generalization ability and adaptability under complex dynamic conditions.

[0109] On the other hand, based on the same inventive concept as the above embodiments, this embodiment also provides a dynamic prediction system for the amount of coking in supercritical aviation kerosene cracking in pipelines. The method of using this system and the dynamic prediction method for the amount of coking in supercritical aviation kerosene cracking in pipelines provided in the above embodiments 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 construct a pyrolysis and coking dataset by combining a 24-step pyrolysis reaction model and a coking kinetics model for aviation kerosene.

[0111] The architecture design module is used to introduce feedback structures into the topology of neural networks to build a time-series prediction model for decomposition and coagulation.

[0112] The model optimization module is used to train and optimize the time-series prediction model based on the pyrolysis and coking dataset;

[0113] The accuracy prediction module is used to test the prediction accuracy of the optimized time-series prediction model based on the pyrolysis and coking dataset.

[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 realize the various steps of the dynamic prediction method for coking amount of supercritical aviation kerosene cracking in pipelines.

[0115] This embodiment also provides a readable storage medium storing a computer program, which, when executed by a processor, implements the various steps of a dynamic prediction method for the amount of coking in supercritical aviation kerosene cracking within a pipeline.

[0116] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for dynamically predicting the amount of supercritical aviation kerosene cracking coking in a pipeline, characterized in that, The method comprises the following steps: A 24-step cracking reaction model and a coking kinetics model of aviation kerosene are combined to construct a cracking coking data set; A feedback structure is introduced into the topology structure of a neural network to construct a time series prediction model for cracking coking; The time series prediction model is trained and optimized based on the cracking coking data set; The amount of supercritical aviation kerosene cracking coking in a pipeline is dynamically predicted based on the optimized time series prediction model; The process of constructing the cracking coking data set in combination with the 24-step cracking reaction model and the coking kinetics model of aviation kerosene comprises: The 24-step cracking reaction model and the coking kinetics model of aviation kerosene are embedded into a solver of computational fluid dynamics to perform transient simulation on different working conditions, and corresponding transient data are generated; time series data of key variables are extracted from the transient data to generate flow, reaction and coking data under multiple transient working conditions, and an initial cracking coking data set is constructed; the initial cracking coking data set is subjected to data cleaning and correction processing; the processed initial cracking coking data set is subjected to format conversion, and each data is labeled with corresponding working condition and time point to obtain a final cracking coking data set; The process of constructing the time series prediction model for cracking coking by introducing a feedback structure into the topology structure of a neural network comprises: A cycle layer is added to the hidden layer of the neural network, and the values of the hidden layer and the output layer after the cycle layer are represented as: S t = f(U X t + W S t-1 ) Y t = g(V · S t ) where S t represents the value of the hidden layer at time t, f represents a transfer function between the values of the input layer and the recurrent layer to the hidden layer, U represents a weight matrix between the output layer to the hidden layer, g represents a transfer function between the values of the hidden layer to the output layer, X t represents the value of the input layer at time t, W represents a weight matrix of the previous value of the hidden layer as the current input, 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, V represents a weight matrix between the hidden layer to the output layer.

2. The dynamic prediction method of the amount of supercritical aviation kerosene cracking coking in a pipeline according to claim 1, characterized in that, Before the time series prediction model is trained based on the cracking coking data set, the method further comprises: The cracking coking data set is subjected to maximum minimum normalization processing, and the processed cracking coking data set is randomly divided into a training set, a validation set and a test set.

3. The dynamic prediction method of the amount of supercritical aviation kerosene cracking coking in a pipeline according to claim 2, characterized in that, The process of training the time series prediction model based on the cracking coking data set comprises: iteratively processing the time series prediction model based on the training set in the cracking coking data set, and calculating the activation values of the input layer, the hidden layer, the output layer and the cycle layer of the time series prediction model, respectively; when the activation values of each layer meet the iteration requirements, the training is completed.

4. The dynamic prediction method of the amount of supercritical aviation kerosene cracking coking in a pipeline according to claim 3, characterized in that, The process of optimizing the time series prediction model based on the cracking coking data set comprises: Based on the validation set in the cracking coking data set, the mean square error and the relative error of the actual output and the expected output of the time series prediction model are calculated; if the mean square error and the relative error do not meet the set value, the topology structure of the time series prediction model is optimized; wherein, the topology structure to be optimized includes the number of layers of the network, the number of nodes of each layer of the network, the activation function, the learning algorithm, the learning rate and the loss function.

5. A system for dynamic prediction of the amount of supercritical aviation kerosene cracking coking in a pipeline, characterized by, The method for dynamically predicting the coking amount of supercritical aviation kerosene cracking in a pipeline according to any one of claims 1-4 comprises a data acquisition module, an architecture design module, a model optimization module, and a precision prediction module. The data acquisition module is used to combine a 24-step cracking reaction model and a coking kinetics model of aviation kerosene to construct a cracking and coking data set. The architecture design module is used to introduce a feedback structure into the topology structure of a 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 and coking data set. The precision prediction module is used to test the prediction accuracy of the optimized time series prediction model based on the cracking and coking data set.

6. An electronic device, comprising: It comprises: a memory and a processor; the memory is used to store a program; the processor is used to execute the program to realize the steps of the method for dynamically predicting the coking amount of supercritical aviation kerosene cracking in a pipeline according to any one of claims 1-4.

7. A readable storage medium, having stored thereon a computer program, characterized in that, When the computer program is executed by the processor, the steps of the method for dynamically predicting the coking amount of supercritical aviation kerosene cracking in a pipeline according to any one of claims 1-4 are realized.

Images