Combustion Simulation and Analysis Method and System Based on ADN-based Liquid Propellant

Abstract

The present invention provides a combustion simulation and analysis method and system based on an ADN-based liquid propellant, which uses a joint learning network for predicting combustion parameters. During network calibration, X single-parameter prediction networks are obtained, and the first inference layers corresponding to the component sets in the single-parameter prediction networks are calibrated. The component sets at the a-th level in the X single-parameter prediction networks are fused to obtain a combustion parameter prediction network, and X second inference layers are set for the a-th fusion component set; the true combustion parameters corresponding to the combustion condition training data of the ADN-based liquid propellant samples are used as prior labels, the predicted combustion parameters of the first inference layer are used as prior labels for the predicted combustion parameters of the second inference layer, and the predicted combustion parameters of the single-parameter prediction network are used as prior labels to calibrate the combustion parameter prediction network. The present invention can improve the accuracy of combustion parameter prediction of the combustion parameter prediction network, so as to improve the accuracy of combustion simulation.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and in particular, to a combustion simulation and analysis method and system based on an ADN-based liquid propellant. Background Art

[0002] In fields such as aerospace, ADN-based liquid propellants are widely used due to their excellent performance. It is crucial to accurately simulate and analyze the combustion characteristics of ADN-based liquid propellants, which helps to deeply understand the combustion process of the propellants, optimize engine design, and improve the performance and reliability of the propulsion system. Currently, in the simulation and analysis of propellant combustion, although some machine learning-based prediction methods have improved the prediction efficiency to a certain extent, due to the unreasonable design of the network structure and the lack of effective tuning strategies, the prediction accuracy is limited. For example, most of the existing prediction networks are designed for a single combustion parameter and cannot accurately predict multiple combustion parameters simultaneously; moreover, during the network training process, the correlations between different parameters and the advantages of existing models are not fully utilized, resulting in a large deviation between the prediction results and the actual combustion situation, making it difficult to meet the requirements of actual engineering applications. Summary of the Invention

[0003] The purpose of the present invention is to provide a combustion simulation and analysis method and system based on an ADN-based liquid propellant. The embodiments of the present invention are implemented as follows:

[0004] In a first aspect, an embodiment of the present invention provides a combustion simulation and analysis method based on an ADN-based liquid propellant, the method including: in response to a simulation analysis instruction, obtaining combustion condition data of a target ADN-based liquid propellant; loading the combustion condition data of the target ADN-based liquid propellant into a pre-tuned combustion parameter prediction network, the combustion parameter prediction network being a joint learning network including a plurality of different branch parameter prediction networks, and one branch parameter prediction network being used to predict one combustion parameter; outputting at least one combustion parameter through the combustion parameter prediction network; and performing visualization processing based on the at least one combustion parameter to obtain a simulation result of the target ADN-based liquid propellant.

[0005] In a second aspect, an embodiment of the present invention provides a computer system, including: one or more processors; a memory; one or more computer programs; wherein the one or more computer programs are stored in the memory and configured to be executed by the one or more processors, and when the one or more computer programs are executed by the processors, the method as described above is implemented.

[0006] In the calibration process of the combustion parameter prediction network of the present invention, a first inference layer is configured on the component set in the single-parameter prediction network for calibration to obtain the predicted combustion parameters of the component set in the single-parameter prediction network, which are used for prior marking of the second inference layer of the component set in the combustion parameter prediction network, strengthening the prediction performance of the component set in the combustion parameter prediction network, so that the component set in the combustion parameter prediction network can have the prediction performance of the single-parameter prediction network during prediction, and improving the parameter prediction accuracy of the combustion parameter prediction network. Further, the present invention determines the predicted combustion parameters of the single-parameter prediction network as the prior marking of the predicted combustion parameters of the combustion parameter prediction network, and aligns the predicted combustion parameters of the single-parameter prediction network with those of the combustion parameter prediction network to improve the accuracy of the predicted combustion parameters of the combustion parameter prediction network. In addition, taking the true combustion parameters corresponding to the combustion condition training data of the ADN-based liquid propellant sample as the prior marking increases the parameter prediction accuracy of the combustion parameter prediction network and helps to perform accurate simulation analysis.

[0007] In the following description, other features will be partly stated. When examining the following content and the drawings, those skilled in the art will partly discover these features, or may learn these features through production or application. By practicing or using various aspects of the methods, tools, and combinations listed in the detailed examples described later, the features in the current application can be implemented and obtained. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] To more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments of the present invention.

[0009] Figure 1 is a flowchart of a combustion simulation analysis method based on an ADN-based liquid propellant provided by an embodiment of the present invention.

[0010] Figure 2 is a schematic diagram of the calibration process of the combustion parameter prediction network provided by an embodiment of the present invention.

[0011] Figure 3 is a schematic diagram of the composition of a computer system provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0012] The following describes the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. The terms used in the implementation part of the embodiments of the present invention are only used to explain the specific embodiments of the present invention, and are not intended to limit the present invention.

[0013] In an embodiment of the present invention, the execution subject of the combustion simulation and analysis method based on ADN-based liquid propellant is a computer system, including but not limited to servers, personal computers, laptop computers, tablet computers, smart phones, etc.

[0014] As Figure 1 shown, the method includes:

[0015] Step 100: In response to the simulation analysis instruction, obtain the combustion condition data of the target ADN-based liquid propellant.

[0016] The computer system is in a waiting state, ready to receive the externally input simulation analysis instruction at any time. This instruction can come from a triggering operation performed by the user on a specific operation interface. For example, when the user clicks the "Start Simulation Analysis" button in the interface of a professional propellant combustion simulation software, the software converts this operation into a digital signal that the computer system can recognize, and then generates a simulation analysis instruction and transmits it to the computer system. Then, after receiving the simulation analysis instruction, the computer system begins to obtain the combustion condition data of the target ADN-based liquid propellant. The combustion condition data contains multiple key elements, which are crucial for accurately simulating the combustion process.

[0017] Among them, temperature is an important data of combustion conditions. Different temperature environments will greatly affect the combustion reaction rate and degree of ADN-based liquid propellants. For example, at a lower initial temperature, the activity of propellant molecules is relatively low, and the combustion reaction may be relatively slow; while at a higher initial temperature, the molecular motion intensifies, and the combustion reaction will be more intense and rapid. The computer system can obtain temperature data by interacting with temperature sensor devices, which are accurately placed at appropriate positions in the simulation environment, monitor in real time and transmit temperature information to the computer system. Pressure is also an indispensable data of combustion conditions. The pressure change in the combustion chamber will directly affect the combustion stability and energy release efficiency of the propellant. For example, a higher pressure environment can promote the collision between propellant molecules and accelerate the progress of the combustion reaction; while unstable pressure may lead to fluctuations or even flameout in the combustion process. The computer system can obtain pressure data by connecting pressure sensors, which can accurately measure the pressure values at different positions and times in the combustion chamber and convert them into digital signals and send them to the computer system. The concentration of the propellant is also key data. The combustion characteristics of ADN-based liquid propellants with different concentrations will be different. When the concentration is higher, the combustion reaction may be more intense and the released energy is also greater; when the concentration is lower, the combustion reaction may be relatively mild. The computer system can obtain the concentration data of the propellant through chemical analysis instruments, which can accurately analyze the propellant sample and transmit the concentration information to the computer system. In addition to the above common combustion condition data, there may also be data such as the injection speed and injection angle of the propellant. The injection speed determines the rate at which the propellant enters the combustion chamber, thereby affecting the intensity and duration of combustion; the injection angle will affect the distribution of the propellant in the combustion chamber and have an impact on the uniformity of combustion. The computer system can obtain these data by connecting relevant flow sensors and angle measurement devices.

[0018] In the actual process of obtaining combustion condition data, the computer system has corresponding interfaces and driver programs to achieve communication with various sensors and measurement devices. For example, for temperature sensors, the computer system needs to interact with the sensors through specific communication protocols (such as SPI, I2C, etc.), receive temperature data according to the format specified by the protocol, and perform corresponding parsing and processing. For devices such as pressure sensors and concentration analysis instruments, appropriate interfaces and driver programs also need to be configured to ensure the accurate acquisition and transmission of data.

[0019] When a computer system obtains the combustion condition data of a target ADN-based liquid propellant, it classifies, stores, and manages these data so that they can be conveniently and quickly retrieved later. For example, temperature data is stored in a dedicated temperature data file, pressure data is stored in a pressure data file, and corresponding indexes are established for each data file to quickly locate and retrieve the data.

[0020] Step 200: Load the combustion condition data of the target ADN-based liquid propellant into a pre-tuned combustion parameter prediction network. The combustion parameter prediction network is a joint learning network, including multiple different branch parameter prediction networks. One branch parameter prediction network is used to predict one combustion parameter.

[0021] The combustion parameter prediction network is a joint learning network, that is, a neural network model that performs multiple tasks. The advantage of this network structure is that it can handle multiple related tasks simultaneously. In the embodiments of the present invention, it is to predict multiple different combustion parameters. Taking the simulation of a rocket engine combustion chamber as an example, during actual operation, it is not only necessary to know the temperature distribution in the combustion chamber, which is crucial for judging whether the combustion chamber wall will be damaged due to high temperature; it is also necessary to understand the pressure change, because the stability of the pressure is directly related to the working stability of the engine; at the same time, the component concentration distribution can help analyze the combustion efficiency of the propellant and the product composition; the combustion efficiency and thrust performance are even the key indicators for measuring the performance of the engine. And the combustion parameter prediction network can integrate these tasks and predict these combustion parameters respectively through different branch parameter prediction networks.

[0022] Specifically, the combustion parameter prediction network contains multiple different branch parameter prediction networks. Each branch parameter prediction network is designed to specifically predict a particular combustion parameter. For example, one of the branch parameter prediction networks focuses on predicting the temperature distribution in the combustion chamber. This branch parameter prediction network will receive combustion condition data related to temperature as input, such as the initial temperature of the propellant, the initial pressure in the combustion chamber, etc. data, and then through its internal network structure and algorithms, predict the temperature distribution at different positions and different times in the combustion chamber. Another example is that another branch parameter prediction network is responsible for predicting the pressure change. It will, based on combustion condition data such as the flow rate of the propellant, the injection speed, etc., after a series of calculations and processes, output the change of the pressure in the combustion chamber over time and space.

[0023] During the process of loading data, the computer system operates strictly in accordance with the input requirements of the combustion parameter prediction network. Different branch parameter prediction networks may have specific requirements for the format and dimensions of the input data. For example, the branch parameter prediction network used to predict the temperature distribution may require the input data to be a multi-dimensional array, where each dimension represents different physical quantities (such as time, spatial coordinates, etc.). The computer system sorts out and converts the combustion condition data according to these requirements to ensure that the data can be correctly input into the corresponding branch parameter prediction network.

[0024] The pre-tuned combustion parameter prediction network is obtained through a large amount of training and optimization. During the training process, a large amount of combustion condition training data of ADN-based liquid propellant samples and the corresponding real combustion parameters are used. By continuously adjusting the weights and parameters of the network, the network can predict the combustion parameters as accurately as possible. For example, when training the branch parameter prediction network for predicting combustion efficiency, a large amount of propellant sample data under different combustion conditions is input into the network, and the actual combustion efficiency of these samples is used as the real label. Through continuous learning and adjustment, the network gradually improves the prediction accuracy of combustion efficiency.

[0025] After the computer system successfully loads the combustion condition data of the target ADN-based liquid propellant into the combustion parameter prediction network, each branch parameter prediction network will start to work. Taking the branch parameter prediction network for predicting the component concentration distribution in the combustion chamber as an example, it will analyze and process the loaded combustion condition data. First, the data is received at the input layer of the network and then sequentially undergoes calculations and conversions by multiple neurons in the hidden layer. Each neuron performs a weighted sum of the input data and undergoes a non-linear transformation through an activation function. Commonly used activation functions such as the ReLU function, and its expression is: f(x)=max(0,x). After a series of calculations and conversions, the prediction result of the component concentration distribution in the combustion chamber is finally obtained at the output layer.

[0026] During the entire loading and prediction process, the computer system monitors the data flow and the running state of the network in real time. If there are data loading errors or abnormal network operations, the computer system issues an alarm in a timely manner and performs corresponding processing. For example, if it is found that the data format does not meet the requirements during data loading, the computer will prompt the user to check the data format and provide corresponding correction suggestions.

[0027] The computer system executes step 200, which loads the combustion condition data of the target ADN-based liquid propellant into the pre-tuned combustion parameter prediction network, which is a key link in the entire combustion simulation analysis process. Through reasonable hardware configuration, effective data loading and preprocessing technical means, as well as accurate network input and operation management, it is ensured that each branch parameter prediction network can accurately predict different combustion parameters, laying a solid foundation for obtaining accurate simulation results subsequently.

[0028] Step 300: Output at least one combustion parameter through the combustion parameter prediction network.

[0029] The combustion parameter prediction network, as a joint learning network, contains multiple different branch parameter prediction networks, and each branch parameter prediction network predicts a specific combustion parameter. The computer system uses the neurons and connection weights in each layer of the network to implement this prediction process. When the combustion condition data enters the combustion parameter prediction network, it first reaches the input layer. The neurons in the input layer are responsible for receiving this data and passing it to the next layer, namely the hidden layer. The hidden layer is the key part of the network for complex calculations and feature extraction. In the hidden layer, the neurons perform weighted summation operations on the input data. After weighted summation, the data will undergo a non-linear transformation through an activation function. Through the activation function, the network can learn the non-linear relationships in the data and enhance the network's expressive ability. After multiple calculations and transformations in the hidden layers, the data finally reaches the output layer. The neurons in the output layer will calculate and output the predicted combustion parameters based on the results passed from the hidden layer.

[0030] For example, for the branch parameter prediction network that predicts the temperature distribution in the combustion chamber, the computer system inputs the combustion condition data related to temperature into this branch network. After the input layer receives the data, the neurons in the hidden layer perform weighted summation and non-linear transformation on the data. In this process, the network will learn the complex relationship between the combustion conditions and the temperature distribution. Finally, the output layer outputs the predicted values of the temperature distribution at different positions and times in the combustion chamber based on these calculation results. After the computer system obtains this temperature distribution data, it outputs it as a combustion parameter. Another example is for the branch parameter prediction network responsible for predicting the combustion efficiency. The computer system also inputs the relevant combustion condition data into this network. After the operations of each layer of the network, the output layer will give the predicted value of the combustion efficiency. The computer system outputs this combustion efficiency value as another combustion parameter.

[0031] Step 400: Perform visualization processing based on at least one combustion parameter to obtain the simulation result of the target ADN-based liquid propellant.

[0032] After the combustion parameter prediction network outputs at least one combustion parameter, the computer system proceeds to step 400 to visualize these combustion parameters, thereby obtaining simulation results for the target ADN-based liquid propellant. Visualization presents abstract combustion parameters in intuitive graphics, images, or animations, enabling users to more clearly understand and analyze the propellant's combustion.

[0033] When performing visualization, the computer system first selects an appropriate visualization method based on the different types of combustion parameters. For the combustion parameter temperature distribution within the combustion chamber, the computer system may visualize it using a two-dimensional or three-dimensional heat map. For example, assuming the space within the combustion chamber can be simplified to a two-dimensional plane, the computer system divides this plane into several grid cells, each corresponding to a specific location. Based on the previously output temperature distribution combustion parameters, each grid cell is assigned a corresponding temperature value. Then, using a color mapping method, different temperature values are mapped to different colors. For example, low-temperature areas can be represented by blue, high-temperature areas by red, and intermediate temperature areas by a gradual transition of colors. By observing the generated two-dimensional heat map, the user can intuitively see the temperature distribution within the combustion chamber, quickly identify high and low temperature areas, and determine whether the combustion is uniform.

[0034] For the combustion parameter of pressure change, the computer system can visualize it using a line graph or bar chart. For example, a line graph uses time as the horizontal axis and pressure as the vertical axis. Based on the output pressure-variation combustion parameter, the computer system plots the corresponding pressure values at different time points in the coordinate system and connects these points to form a broken line. This allows the user to clearly understand the pressure change trend over time within the combustion chamber by observing the line, and to determine whether the pressure is stable or experiencing excessive pressure fluctuations.

[0035] For the combustion parameter of component concentration distribution, the computer system can use contour plots or three-dimensional stereograms. When using contour plots, the computer system divides the space within the combustion chamber into specific areas and calculates the component concentration values at different locations based on the combustion parameters of the component concentration distribution. Points with the same concentration value are then connected to form contour lines. These contour lines can intuitively demonstrate the distribution pattern of component concentrations, and users can determine the gradient of concentration changes by the density of the contour lines.

[0036] In the process of visualization, the computer system needs to use some technical means to achieve it. In terms of data processing, the computer system performs necessary preprocessing on the output combustion parameters. For example, the data is smoothed to reduce the noise interference in the data. Assume that the combustion parameter data sequence is , the moving average method is used for smoothing, and the new data sequence The calculation formula is: (where k is the smoothing window size, is the floor function). The data after smoothing can more accurately reflect the change trend of combustion parameters, providing a more reliable data basis for subsequent visualization.

[0037] In terms of graph drawing, the computer system calls a professional visualization library or tool. For example, the Matplotlib library in Python is a widely used visualization tool that provides rich drawing functions and methods. The computer system can use the functions in the Matplotlib library to draw basic graphs such as line charts and bar charts. For the drawing of more complex three-dimensional graphs, the computer system can rely on libraries such as Mayavi, which can visualize three-dimensional space data and present the combustion parameters in an intuitive three-dimensional graph.

[0038] In terms of color mapping, the computer system selects a suitable color mapping scheme according to the range and characteristics of the combustion parameters. Common color mapping schemes include Jet, Viridis, etc. Taking the Viridis color mapping as an example, it is a color mapping widely used in scientific visualization, with good perceptual uniformity and color-blind friendliness. The computer system maps the minimum and maximum values of the combustion parameters to the color space of the Viridis color mapping, so that different combustion parameter values correspond to different colors, thus achieving an intuitive visualization display.

[0039] When performing visualization processing, the computer system also considers the layout and annotation of the graph. For example, adding axis labels, titles, legends and other information to the drawn graph so that users can better understand the meaning represented by the graph. For different combustion parameter graphs, the computer system reasonably arranges their layout on the display interface to make the entire visualization result clearer and more beautiful. By visualizing the combustion parameters, the computer system finally obtains the simulation results of the target ADN-based liquid propellant. These simulation results are presented in the form of intuitive graphs, images or animations. Users can observe these visualization results to deeply understand the combustion characteristics of the ADN-based liquid propellant under different combustion conditions. For example, by observing the visualization results of the temperature distribution, users can judge whether the combustion of the propellant is sufficient and whether there is local overheating; by analyzing the visualization graph of the pressure change, users can evaluate the stability of the combustion process and provide a basis for optimizing the combustion system.

[0040] Please refer to Figure 2 , in an implementation scheme, the combustion parameter prediction network is calibrated through the following steps:

[0041] Step 10: Obtain X single-parameter prediction networks. Different single-parameter prediction networks are used to predict different combustion parameters. The set of components at the same level in different single-parameter prediction networks has the same network architecture, where X is a positive integer greater than 1.

[0042] When the computer system starts to calibrate the combustion parameter prediction network, it first executes Step 10 to obtain X single-parameter prediction networks. Here, the single-parameter prediction network is a network model specifically for predicting a particular combustion parameter.

[0043] Different single-parameter prediction networks are used to predict different combustion parameters. For example, in the combustion simulation scenario of ADN-based liquid propellants, there are multiple combustion parameters to be predicted, such as the temperature distribution, pressure change, and component concentration distribution in the combustion chamber. The computer system constructs corresponding single-parameter prediction networks for each different combustion parameter. Taking the temperature distribution prediction as an example, a dedicated single-parameter prediction network will be constructed. The input data of this network will revolve around factors related to temperature, such as the initial temperature of the propellant and the chemical reaction heat during combustion. Its output is the predicted temperature values at various positions and different times in the combustion chamber. Similarly, for pressure change prediction, there will be another single-parameter prediction network that receives input data related to pressure, such as propellant flow rate and injection speed, and then outputs the predicted values of the pressure change in the combustion chamber over time and space.

[0044] The set of components at the same level in different single-parameter prediction networks has the same network architecture. This means that when constructing these single-parameter prediction networks, the computer system follows certain design rules, making the structures of each network similar at the same level. For example, each single-parameter prediction network may include an input layer, multiple hidden layers, and an output layer. At the input layer, whether it is the temperature prediction network or the pressure prediction network, it is responsible for receiving their respective relevant input data and passing it to the next layer. The set of components in the hidden layer, such as the number of neurons and connection methods, is the same at the same level. Assuming that each single-parameter prediction network has three hidden layers, then in the first hidden layer, the number of neurons, the choice of activation function, etc. of each network are consistent. This consistency helps the subsequent network fusion and calibration process.

[0045] The computer system can obtain these X single-parameter prediction networks through various technical means. In terms of constructing the network structure, the computer system can use common deep learning frameworks such as TensorFlow or PyTorch. Taking TensorFlow as an example, the computer system can utilize various functions and classes it provides to define the hierarchical structure of the network. For example, use `tf.keras.Sequential` to build a sequential model, and define the hidden layer and output layer by adding `tf.keras.layers.Dense` layers. When determining the input and output of each single-parameter prediction network, the computer system makes settings according to the characteristics of different combustion parameters. For the temperature prediction network, the computer system analyzes the physical factors related to temperature, determines the appropriate input data type and dimension, and determines the number of neurons in the output layer and the output format according to the accuracy requirements of temperature prediction.

[0046] When training these single-parameter prediction networks, the computer system uses a large amount of ADN-based liquid propellant sample data and the corresponding real combustion parameters as the training set. For example, for the temperature prediction network, propellant sample data under different combustion conditions will be collected, including information such as the composition of the propellant and the injection method as input, and at the same time, the temperature distribution in the combustion chamber actually measured for these samples will be used as the true label. The computer system adjusts the weight parameters of the network through the backpropagation algorithm to minimize the error between the predicted value and the true value. The core of the backpropagation algorithm is to calculate the gradient of the loss function with respect to the network weights, and its calculation formula is: , where represents the gradient of the loss function L with respect to the weight w. By continuously iteratively updating the weights, the single-parameter prediction network can gradually learn the relationship between the input data and the target combustion parameters, thereby improving the prediction accuracy.

[0047] Step 20: Calibrate the first inference layer corresponding to each component set in the single-parameter prediction network. The first inference layer is used to obtain the predicted combustion parameters based on the output of the component set.

[0048] After the computer system obtains the X single-parameter prediction networks, it calibrates the first inference layer corresponding to each component set in these networks. The first inference layer plays a key role in the single-parameter prediction network. It is responsible for obtaining the predicted combustion parameters based on the output of the component set.

[0049] Taking a single-parameter prediction network for predicting the temperature distribution in a combustion chamber as an example, after a series of data processing and feature extraction on the component set in this network, the results will be passed to the corresponding first inference layer. The component set can be understood as a combination of neurons at a certain level in the network, and they jointly process the input data. For example, in this temperature prediction network, the component set may be a certain layer in the hidden layer, which calculates and transforms the input combustion condition data related to temperature, such as the initial temperature of the propellant, the heat of chemical reaction, etc., to extract features that have an important impact on the temperature distribution.

[0050] After receiving the output of the component set, the first inference layer will perform inference calculations based on these outputs to obtain the predicted combustion parameters, that is, the predicted values of the temperature distribution in the combustion chamber. The computer system tunes the first inference layer to enable it to make more accurate and reasonable predictions based on the output of the component set.

[0051] During the tuning process, the computer system adopts appropriate technical means. First, in terms of data, the computer system uses a large amount of training data with real temperature distribution labels. These data are obtained from actual experiments or simulations and contain the temperature distribution in the combustion chamber under various different combustion conditions. The computer system inputs these data into the single-parameter prediction network, allows the component set to process them, and then the first inference layer makes a temperature distribution prediction based on the output of the component set.

[0052] Next, the computer system calculates the prediction error. Assuming that the temperature distribution predicted by the first inference layer is , and the actual real temperature distribution is T, then the error calculation method can adopt the mean square error (MSE), and its formula is: , where n represents the number of data samples, are the predicted temperature and the real temperature of the i-th sample respectively. Through this formula, the computer system can quantify the difference between the predicted value and the real value of the first inference layer.

[0053] Then, the computer system tunes the first inference layer according to the calculated error. This is usually achieved through the backpropagation algorithm. The backpropagation algorithm will propagate the error signal from the output layer back to the input layer, calculate the gradients of the error with respect to various parameters (such as weights and biases) in the network. The computer system adjusts the parameters of the first inference layer according to these gradients to gradually reduce the error.

[0054] In actual operation, the computer system repeats the above process multiple times, using different training data samples to calibrate the first inference layer. As the calibration progresses, the prediction accuracy of the first inference layer will continuously improve. For example, before the calibration starts, there may be a large deviation between the temperature distribution predicted by the first inference layer and the true temperature distribution. After multiple calibrations, the prediction results will gradually approach the true value, and the mean square error will continuously decrease.

[0055] Step 30: Fuse the component sets at the a-th level in the X single-parameter prediction networks to obtain a combustion parameter prediction network. The combustion parameter prediction network corresponds to X parameter prediction branches, and the combustion parameter prediction network includes the first fusion component set to the a-th fusion component set obtained by fusion, where a is a positive integer greater than 0.

[0056] Step 30 constructs a combustion parameter prediction network by fusing the component sets at the a-th level in the X single-parameter prediction networks. This process is crucial for integrating the advantages of multiple single-parameter prediction networks and improving the overall network's prediction ability for multiple combustion parameters.

[0057] Taking X = 3 as an example, assume that these three single-parameter prediction networks are respectively used to predict the temperature distribution, pressure change, and component concentration distribution in the combustion chamber. At a specific level, that is, the a-th level, each single-parameter prediction network has a corresponding component set. For example, for the single-parameter network for temperature prediction, the component set at the a-th level has extracted and processed the features related to temperature through the operations of the previous levels; the component set at the a-th level of the single-parameter network for pressure prediction has similarly processed the information related to pressure; the component set at the a-th level of the single-parameter network for component concentration prediction has also completed the processing of the data related to concentration.

[0058] The computer system fuses the component sets at the a-th level in these three single-parameter prediction networks. The fusion process is not a simple superposition but comprehensively considers the output information of each component set to construct a new component set that can simultaneously process the features related to multiple combustion parameter predictions. Specifically, the computer system adopts a suitable fusion strategy. A common method is to perform a concatenation operation on the outputs of each component set. Assume that the output of the component set at the a-th level of the temperature prediction network is a vector T of dimension m a , the output of the component set at the a-th level of the pressure prediction network is a vector P of dimension n a , and the output of the component set at the a-th level of the component concentration prediction network is a vector C of dimension p a . Then the fused vector F a can be expressed as F a =[T a , P a , C a , and its dimension is m + n + p.

[0059] In this way, the computer system sequentially performs fusion operations on the component sets at the a-th level in X single-parameter prediction networks, starting from the first level, gradually constructs a fused component set, and finally obtains a combustion parameter prediction network. This combustion parameter prediction network corresponds to X parameter prediction branches, which are respectively responsible for predicting different combustion parameters, namely, the three branches of temperature distribution, pressure change, and component concentration distribution.

[0060] The combustion parameter prediction network includes the first fused component set to the a-th fused component set obtained by fusion. These fused component sets inherit some characteristics of the corresponding-level component sets in the single-parameter prediction networks, and at the same time integrate the information of multiple single-parameter networks, enabling the network to analyze and predict combustion parameters from multiple perspectives. For example, the first fused component set fuses the outputs of the first-level component sets of each single-parameter prediction network, and can comprehensively consider the influence of various combustion condition factors on combustion parameters; as the level increases, subsequent fused component sets continuously dig deeper and integrate more complex features, providing richer information for accurately predicting combustion parameters.

[0061] Step 40: Set X second inference layers for the a-th fused component set, and different second inference layers are used to obtain the predicted combustion parameters corresponding to different parameter prediction branches based on the output of the fused component set.

[0062] After the computer system completes the fusion of the component sets at the a-th level in X single-parameter prediction networks and obtains a combustion parameter prediction network containing the a-th fused component set, it then proceeds to set X second inference layers for the a-th fused component set. The purpose of this operation is to enable the combustion parameter prediction network to accurately obtain the corresponding predicted combustion parameters for different parameter prediction branches based on the output of the fused component set.

[0063] Taking a specific example to illustrate, assume X = 4, that is, there are four single-parameter prediction networks, which are respectively used to predict the temperature distribution, pressure change, combustion efficiency, and component concentration in the combustion chamber. After step 30, the computer system has fused the component sets at the a-th level in these four single-parameter prediction networks to form the a-th fused component set. At this time, in order to enable the combustion parameter prediction network to predict these four combustion parameters respectively, the computer system sets four second inference layers for the a-th fused component set.

[0064] Each second inference layer has a specific function to obtain the predicted combustion parameters corresponding to different parameter prediction branches based on the output of the set of fusion components. For the temperature distribution parameter prediction branch, the corresponding second inference layer analyzes and processes the output of the a-th set of fusion components. The output of the set of fusion components contains the comprehensive information from multiple single-parameter prediction networks at the a-th level. This second inference layer extracts temperature-related features from it and obtains the predicted value of the temperature distribution through a specific calculation method. For example, this second inference layer may perform matrix multiplication operations according to some feature vectors in the output of the set of fusion components and in combination with a pre-trained weight matrix. Assuming the output vector of the set of fusion components is O and the weight matrix is W, through the formula T = W × O (where T is the intermediate result related to temperature obtained by preliminary calculation), and then through some non-linear transformations (such as using the ReLU activation function: f(x) = max(0, x)), the predicted value of the temperature distribution in the combustion chamber is finally obtained.

[0065] Similarly, for the second inference layer corresponding to the pressure change parameter prediction branch, it processes the output of the set of fusion components with different weight matrices and calculation logics to obtain the predicted value of the pressure change. This second inference layer focuses on the pressure-related information in the fusion output and through corresponding calculations, such as using another set of weight matrices W' to operate on the fusion output vector O to get P = W' × O (P is the intermediate result related to pressure), and then through appropriate transformations, the predicted result of the pressure change is obtained.

[0066] For the combustion efficiency and component concentration parameter prediction branches, the corresponding second inference layers also operate on the output of the set of fusion components in their respective ways. The second inference layer of the combustion efficiency prediction branch calculates the predicted value of the combustion efficiency according to the combustion efficiency-related features in the fusion output by using specific algorithms and weights; the second inference layer of the component concentration prediction branch focuses on extracting the component concentration-related information and obtains the predicted value of the component concentration through corresponding calculations.

[0067] The computer system enables the combustion parameter prediction network to accurately obtain the corresponding predicted combustion parameters for different parameter prediction branches based on the output of the set of fusion components by setting X second inference layers for the a-th set of fusion components. This step further improves the structure of the combustion parameter prediction network, provides key support for accurately predicting various combustion parameters through this network subsequently, and helps to improve the accuracy and reliability of the entire combustion simulation and analysis.

[0068] Step 50: Calibrate the combustion parameter prediction network by using the true combustion parameters corresponding to the combustion condition training data of the ADN-based liquid propellant sample as prior labels, the predicted combustion parameters of the first inference layer as prior labels for the predicted combustion parameters of the second inference layer, and the predicted combustion parameters of the single-parameter prediction network as prior labels for the predicted combustion parameters of the combustion parameter prediction network.

[0069] After the computer system constructs the combustion parameter prediction network and sets up the structures of each layer in the previous steps, it enters Step 50 to further calibrate the network. This step optimizes the performance of the combustion parameter prediction network by using various prior label information, making its prediction results closer to the true values.

[0070] Take the example that there are three single-parameter prediction networks, which are respectively used to predict the temperature distribution, pressure change, and component concentration distribution in the combustion chamber. For the combustion condition training data of the ADN-based liquid propellant sample, the computer system has obtained the true combustion parameters corresponding to these samples under different combustion conditions. For example, under a series of specific combustion conditions, the true distribution data of the temperature in the combustion chamber, the actual data of the pressure change, and the accurate values of the concentrations of each component are obtained through precise measurement. These true combustion parameters are used by the computer system as important prior labels.

[0071] In each single-parameter prediction network, the first inference layer has been calibrated and can output predicted combustion parameters. For example, the first inference layer of the temperature prediction single-parameter network obtains the predicted value of the temperature distribution in the combustion chamber based on the output of the network component set; the first inference layer of the pressure prediction single-parameter network also outputs the predicted result of the pressure change; the first inference layer of the component concentration prediction single-parameter network also generates the corresponding predicted value. The computer system uses the predicted combustion parameters of these first inference layers as prior labels for the predicted combustion parameters of the second inference layer. This means that in the combustion parameter prediction network, when the second inference layer makes a prediction, it will refer to the relatively reliable prediction results already obtained by the first inference layer and conduct further inference and optimization based on this.

[0072] At the same time, the predicted combustion parameters when the single-parameter prediction networks operate independently are also used by the computer system as prior labels for the predicted combustion parameters of the combustion parameter prediction network. For example, the temperature single-parameter prediction network independently predicts the temperature value of a certain area in the combustion chamber at a certain moment, and this value will be used as a reference for the combustion parameter prediction network when predicting the temperature of the same area and moment.

[0073] Based on these prior labels, the computer system tunes the combustion parameter prediction network. The computer system uses a loss function to measure the difference between the predicted values and the prior labels, such as the mean squared error (MSE) loss function. The computer system adjusts the weights of the combustion parameter prediction network through the backpropagation algorithm according to the errors calculated by these loss functions. The backpropagation algorithm calculates the gradients of the weights of each layer of the network based on the errors to update the weights and gradually reduce the errors.

[0074] During the tuning process, the computer system continuously compares the predicted values of the combustion parameter prediction network with various prior labels. For example, it compares the temperature distribution predicted by the combustion parameter prediction network with the true temperature distribution, the temperature distribution predicted by the first inference layer, and the temperature distribution predicted by the single-parameter prediction network. According to the errors generated by the comparison, the weights of the network are continuously adjusted. After multiple iterations, the combustion parameter prediction network will gradually learn how to better utilize the information of these prior labels, thereby improving the prediction accuracy.

[0075] By using the true combustion parameters corresponding to the training data of the combustion conditions of the ADN-based liquid propellant samples, the predicted combustion parameters of the first inference layer, and the predicted combustion parameters of the single-parameter prediction network as prior labels, the computer system uses technical means such as loss functions and backpropagation algorithms to tune the combustion parameter prediction network, enabling the network to more accurately predict combustion parameters, improving the performance and reliability of the network, and providing more accurate results for subsequent combustion simulation and analysis based on this network.

[0076] In one implementation, the architecture of the a-th fusion component set has the same network architecture as the component set at the a-th level in the single-parameter prediction network, and the initial network parameter variables of the a-th fusion component set are the average values of the network parameter variables of the component sets at the a-th level in X single-parameter prediction networks.

[0077] When the computer system constructs the combustion parameter prediction network, it involves setting the architecture of the a-th fusion component set. It is stipulated that the architecture of the a-th fusion component set has the same network architecture as the component set at the a-th level in the single-parameter prediction network. This means that in the design of the network structure, the two are consistent.

[0078] For example, assume there are three single-parameter prediction networks, which are respectively used to predict the temperature, pressure, and component concentration in the combustion chamber. Each single-parameter prediction network has multiple layers. When a = 2, it is in the second layer. If the single-parameter prediction network adopts a fully connected layer architecture in the second layer, which contains 100 neurons, is connected to the previous layer through a weight matrix, and uses the ReLU activation function to introduce non-linearity, then the a-th fusion component set will also adopt a fully connected layer architecture, also set 100 neurons, and the connection method and the use of the activation function are the same as those of the second layer of the single-parameter prediction network. Moreover, the initial network parameters of the a-th fusion component set are the average of the network parameters of the component sets at the a-th layer in X single-parameter prediction networks.

[0079] In one implementation, in step 50, using the true combustion parameters corresponding to the combustion condition training data of the ADN-based liquid propellant sample as the prior label, using the predicted combustion parameters of the first inference layer as the prior label for the predicted combustion parameters of the second inference layer, and using the predicted combustion parameters of the single-parameter prediction network as the prior label for the predicted combustion parameters of the combustion parameter prediction network to calibrate the combustion parameter prediction network, including:

[0080] Step 51: Determine the parameter prediction error, the first probability distribution error, and the second probability distribution error. The parameter prediction error represents the error between the predicted combustion parameters of the combustion parameter prediction network and the true combustion parameters. The first probability distribution error represents the probability distribution gap between the predicted combustion parameters of the combustion parameter prediction network and the single-parameter prediction network. The second probability distribution error represents the probability distribution gap between the predicted combustion parameters of the first inference layer and the second inference layer;

[0081] Step 52: Calibrate the a-th fusion component set and the second inference layer in the combustion parameter prediction network according to the parameter prediction error, the first probability distribution error, and the second probability distribution error.

[0082] Step 51 is to determine the parameter prediction error, the first probability distribution error, and the second probability distribution error. The parameter prediction error represents the error between the predicted combustion parameters of the combustion parameter prediction network and the true combustion parameters, which intuitively reflects the deviation degree of the network prediction result from the actual situation. For example, when predicting the temperature distribution in the combustion chamber, the predicted temperature value at a certain position output by the combustion parameter prediction network is while the true temperature value at this position obtained through precise measurement is then the parameter prediction error at this point can be expressed as . For the prediction of the temperature distribution in the entire combustion chamber, the computer system calculates this kind of error at all positions and through a suitable statistical method (such as the mean square error, the formula is where N is the total number of measurement positions, to quantify the overall parameter prediction error using the predicted temperature and the true temperature at the i-th position respectively).

[0083] The first probability distribution error represents the difference in the probability distributions of the predicted combustion parameters between the combustion parameter prediction network and the single-parameter prediction network. A probability distribution describes the likelihood of a random variable taking different values. In the prediction of combustion parameters, the prediction results of different prediction networks for the same combustion parameter may have different probability distributions. Taking the prediction of combustion efficiency as an example, the predicted values of combustion efficiency given by the combustion parameter prediction network have a certain probability distribution in different value ranges and the predicted values given by the single-parameter prediction network responsible for predicting combustion efficiency also have corresponding probability distributions . The computer system can use some measurement methods to calculate the difference between these two probability distributions, such as the KL divergence (Kullback-Leibler divergence), and the formula is . The smaller the KL divergence value, the closer the two probability distributions are, and the smaller the first probability distribution error is, which means that the combustion parameter prediction network and the single-parameter prediction network are more similar in the prediction of this combustion parameter.

[0084] The second probability distribution error represents the difference in the probability distributions of the predicted combustion parameters between the first inference layer and the second inference layer. For example, when predicting the pressure change, the first inference layer obtains the predicted value of the pressure change based on the output of the single-parameter prediction network component set, and its probability distribution is ; the second inference layer obtains the predicted value of the pressure change based on the output of the fusion component set, and the probability distribution is . The computer system can also use a method similar to the KL divergence (such as ) to calculate the difference between these two probability distributions to determine the second probability distribution error.

[0085] Step 52 is to calibrate the a-th fusion component set and the second inference layer in the combustion parameter prediction network according to the parameter prediction error, the first probability distribution error, and the second probability distribution error. First, the computer system determines the comprehensive error based on these errors. The determination of the comprehensive error is not a simple addition, but rather takes into account the importance of each error, that is, weighted summation is performed through the error influence coefficient. Suppose the first probability distribution error is , and its first error influence coefficient is ; the second probability distribution error is , and the second error influence coefficient is ; the parameter prediction error is , and the third error influence coefficient is , and . Then the formula for calculating the comprehensive error is 。

[0086] Next, the computer system calibrates the a-th fusion component set and the second inference layer in the combustion parameter prediction network according to the comprehensive error. In deep learning, a common method is to update the network parameters through the backpropagation algorithm. The backpropagation algorithm calculates the gradients of the error with respect to the network parameters (such as weights and biases) based on the comprehensive error. Then, according to the gradient descent method, it controls the step size of weight update. The parameters of the second inference layer are updated in a similar way.

[0087] In actual operation, the computer system repeats the above process multiple times. Each time, three types of errors are calculated, and after determining the comprehensive error, the parameters of the a-th fusion component set and the second inference layer are updated. As the number of iterations increases, the comprehensive error gradually decreases, which means that the prediction performance of the combustion parameter prediction network is continuously improving.

[0088] In one implementation, step 51, determining the parameter prediction error, the first probability distribution error, and the second probability distribution error, includes:

[0089] Step 511: Determine the parameter prediction error based on the predicted combustion parameters of the X parameter prediction branches of the combustion parameter prediction network and the corresponding true combustion parameters of the X parameter prediction branches;

[0090] Step 512: For the same parameter prediction branch, determine the first predicted combustion parameter of the parameter prediction branch by the combustion parameter prediction network, and the second predicted combustion parameter of the parameter prediction branch by the single-parameter prediction network corresponding to the parameter prediction branch; Determine the first probability distribution error based on the probability distributions of the first predicted combustion parameter and the second predicted combustion parameter;

[0091] Step 513: For the same parameter prediction branch, determine the third predicted combustion parameter of the second inference layer corresponding to the parameter prediction branch, and the fourth predicted combustion parameter of the first inference layer corresponding to the parameter prediction branch; Determine the second probability distribution error based on the probability distributions of the third predicted combustion parameter and the fourth predicted combustion parameter.

[0092] The combustion parameter prediction network includes X parameter prediction branches, and each branch is responsible for predicting a specific combustion parameter, such as temperature distribution, pressure change, component concentration distribution, etc. For each parameter prediction branch, the computer system compares the predicted value of the network with the true value obtained from actual measurement.

[0093] Taking the prediction of the temperature distribution in the combustion chamber as an example, assume that the combustion parameter prediction network has 5 parameter prediction branches, and one of the branches is dedicated to temperature prediction. In a simulation, this temperature prediction branch predicts the temperature at 100 different positions in the combustion chamber, obtaining a set of predicted temperature values Meanwhile, through high-precision temperature measurement devices, the true temperature value sets at these 100 positions are obtained. .

[0094] The computer system can use multiple methods to calculate the parameter prediction error. The commonly used one is the mean squared error (MSE) method, and the formula is: , where n is the number of measurement positions. In this example, n = 100. Through this formula, the computer system squares the difference between each predicted temperature value and the corresponding true temperature value, then sums them up and divides by the number of positions to obtain the parameter prediction error of this temperature prediction branch. For example, after calculation, MSE = 25, which means there is a certain deviation between the predicted value and the true value of this temperature prediction branch.

[0095] For other parameter prediction branches, such as the pressure change prediction branch, the component concentration distribution prediction branch, etc., the computer system will also adopt similar methods to calculate their respective parameter prediction errors. In actual operation, the computer system uses the functions provided by programming languages and deep learning frameworks to implement these calculations.

[0096] In step 512, for the same parameter prediction branch, determine the first predicted combustion parameter of the combustion parameter prediction network for this parameter prediction branch, and the second predicted combustion parameter of the single-parameter prediction network corresponding to this parameter prediction branch for this branch; based on the probability distributions of the first predicted combustion parameter and the second predicted combustion parameter, determine the first probability distribution error. Each parameter prediction branch will generate prediction results in the combustion parameter prediction network and the corresponding single-parameter prediction network.

[0097] Continuing with the temperature prediction branch as an example, the combustion parameter prediction network makes predictions for this temperature prediction branch, obtaining a series of predicted temperature values, which form the first predicted combustion parameter set . And the single-parameter prediction network dedicated to temperature prediction makes predictions for the same combustion conditions, obtaining the second predicted combustion parameter set .

[0098] The computer system needs to analyze the probability distribution situations of these two groups of predicted values. First, statistical analysis will be performed on these two groups of data. For example, the temperature range is divided into several intervals, and the frequencies of the predicted values appearing in each interval are counted to obtain their probability distributions. Suppose the temperature range is [200, 1000] degrees Celsius, divided into 10 intervals: [200 - 300), [300 - 400), …, [900 - 1000]. The computer system counts the frequencies of the predicted values of the combustion parameter prediction network in each interval to form a probability distribution ; the frequencies of the predicted values of the single-parameter prediction network in each interval to form a probability distribution 。

[0099] To measure the difference between these two probability distributions, the computer system uses the Kullback-Leibler divergence to calculate the first probability distribution error, and the formula is: . Here, x represents the temperature range, are the probability values of the two probability distributions in the x range respectively. For example, after calculation, we get , and this value reflects the difference degree of the probability distributions of the combustion parameter prediction network and the single-parameter prediction network in temperature prediction, that is, the first probability distribution error.

[0100] For other parameter prediction branches, the computer system will also determine their prediction values in the combustion parameter prediction network and the corresponding single-parameter prediction network respectively according to the above method, analyze their probability distributions, and calculate the first probability distribution error.

[0101] In step 513, for the same parameter prediction branch, determine the third predicted combustion parameter of the second inference layer corresponding to this parameter prediction branch, and the fourth predicted combustion parameter of the first inference layer corresponding to this parameter prediction branch; according to the probability distributions of the third predicted combustion parameter and the fourth predicted combustion parameter, determine the second probability distribution error. In the combustion parameter prediction network, each parameter prediction branch has a corresponding first inference layer and second inference layer, which produce prediction results respectively.

[0102] Still taking the temperature prediction branch as an example, the second inference layer corresponding to this branch performs temperature prediction according to the output of the fusion component set, and obtains the third predicted combustion parameter set ; and the first inference layer corresponding to this branch performs temperature prediction according to the output of the single-parameter prediction network component set, and obtains the fourth predicted combustion parameter set .

[0103] The computer system analyzes the probability distributions of these two groups of prediction values again. Similarly, the temperature range is divided into several intervals, and the frequencies of the third predicted combustion parameter and the fourth predicted combustion parameter appearing in each interval are counted, and the probability distributions are obtained respectively, where y represents the temperature range.

[0104] Then, the computer system uses the Kullback-Leibler divergence to calculate the second probability distribution error, and the formula is: . Suppose that after calculation, we get , and this value reflects the difference in the probability distributions of the second inference layer and the first inference layer in temperature prediction, that is, the second probability distribution error.

[0105] For other parameter prediction branches, the computer system will also repeat this process, respectively determine the predicted values of their corresponding second inference layer and first inference layer, analyze the probability distribution, and calculate the second probability distribution error. Through steps S511 - S513, the computer system comprehensively and meticulously determines the parameter prediction error, the first probability distribution error, and the second probability distribution error. These error metrics reflect the performance of the combustion parameter prediction network and the differences from other related predictions from different perspectives. The parameter prediction error directly measures the deviation degree between the network predicted value and the true value; the first probability distribution error compares the differences in the prediction result distributions between the combustion parameter prediction network and the single-parameter prediction network; the second probability distribution error evaluates the distribution differences in the prediction results of different inference layers within the combustion parameter prediction network. These error information provide precise data support for the subsequent calibration of the combustion parameter prediction network. The computer system can adjust the parameters and structure of the network targeted according to these errors to improve the prediction accuracy and performance of the network, so as to better serve the combustion simulation analysis of ADN-based liquid propellants.

[0106] In one implementation, step 52, according to the parameter prediction error, the first probability distribution error, and the second probability distribution error, for the a-th fusion component set and the second inference layer in the combustion parameter prediction network, includes:

[0107] Step 521: Determine the comprehensive error according to the parameter prediction error, the first probability distribution error, and the second probability distribution error;

[0108] Step 522: Calibrate the a-th fusion component set and the second inference layer in the combustion parameter prediction network according to the comprehensive error.

[0109] In step 521, after the computer system obtains the parameter prediction error, the first probability distribution error, and the second probability distribution error, assume the parameter prediction error is , which directly reflects the deviation degree between the prediction result of the combustion parameter prediction network and the true combustion parameters. This is a key indicator to measure the prediction accuracy of the network and is of great significance for the evaluation of the network performance. Therefore, a relatively large weight is assigned to it, and its third error influence coefficient is set as ; the first probability distribution error is , which reflects the difference in the probability distribution of predicting combustion parameters between the combustion parameter prediction network and the single-parameter prediction network, and helps to analyze the performance of the network when integrating multi-source information. Set its first error influence coefficient as ; the second probability distribution error is , which reflects the difference in the probability distribution of the prediction results of the first inference layer and the second inference layer in the combustion parameter prediction network and is very important for optimizing the internal inference process of the network. Set its second error influence coefficient as , and satisfy 。

[0110] Comprehensive error The calculation formula is as follows: 。

[0111] In step 522, the a-th fusion component set and the second inference layer in the combustion parameter prediction network are calibrated according to the comprehensive error. After the computer system determines the comprehensive error, it will use the backpropagation algorithm to calibrate the a-th fusion component set and the second inference layer in the combustion parameter prediction network. The core idea of the backpropagation algorithm is to calculate the gradients of the parameters (such as weights and biases) in each layer of the network according to the error, and then update the parameters according to these gradients to gradually reduce the error.

[0112] In the actual calibration process, the computer system backpropagates the comprehensive error to each parameter of the a-th fusion component set and the second inference layer. After updating all the parameters once, one iteration is completed. Then, the new parameter prediction error, the first probability distribution error, and the second probability distribution error are calculated again, the comprehensive error is re-determined, and the next round of parameter update continues. As the number of iterations increases, the comprehensive error will gradually decrease. For example, after the first round of iteration, the comprehensive error drops from 0.165 to 0.15; after multiple rounds of iteration, the comprehensive error may be reduced to 0.1 or even lower. This indicates that the parameters of the a-th fusion component set and the second inference layer in the combustion parameter prediction network are continuously adjusted and optimized, and the prediction performance of the network is gradually improved.

[0113] In one implementation, in step 521, determining the comprehensive error according to the parameter prediction error, the first probability distribution error, and the second probability distribution error includes:

[0114] Step 5211: Determine the comprehensive error according to the first probability distribution error, the first error influence coefficient of the first probability distribution error, the second probability distribution error, the second error influence coefficient of the second probability distribution error, the parameter prediction error, and the third error influence coefficient of the parameter prediction error; where the third error influence coefficient is greater than the second error influence coefficient, and the second error influence coefficient is greater than the first error influence coefficient.

[0115] The computer system determines the comprehensive error according to the first probability distribution error, the first error influence coefficient of the first probability distribution error, the second probability distribution error, the second error influence coefficient of the second probability distribution error, the parameter prediction error, and the third error influence coefficient of the parameter prediction error. This process involves multiple error indicators and their corresponding weight coefficients, and through a specific calculation method, it comprehensively measures the performance deviation of the combustion parameter prediction network.

[0116] The first probability distribution error reflects the difference in the probability distributions of the predicted combustion parameters between the combustion parameter prediction network and the single-parameter prediction network, and embodies the degree of consistency of the network in integrating information from different prediction sources. For example, when predicting the pressure change in the combustion chamber, the predicted pressure values obtained by the combustion parameter prediction network exhibit a certain probability distribution in different pressure intervals, and the single-parameter prediction network also has a corresponding probability distribution. The difference between the two is the first probability distribution error. Suppose the first probability distribution error is denoted as E1. The first error influence coefficient of the first probability distribution error is a weight value used to measure the importance of the first probability distribution error in the comprehensive error. This coefficient is set according to the goals of network calibration and the degree of attention to different errors, and is denoted as α. For example, if we hope to focus on the consistency in probability distribution between the combustion parameter prediction network and the single-parameter prediction network, we can appropriately increase the value of α.

[0117] The second probability distribution error refers to the difference in the probability distributions of the predicted combustion parameters between the first inference layer and the second inference layer, which helps to evaluate the stability and coherence of the prediction results at different inference levels within the network. For example, for the prediction of component concentration, the difference in the probability distributions of the predicted values of the first inference layer and the second inference layer in different concentration intervals is the second probability distribution error, denoted as E2.

[0118] The second error influence coefficient of the second probability distribution error is used to determine the weight of the second probability distribution error in the comprehensive error, denoted as β. This coefficient can be adjusted according to the need to optimize the internal inference structure of the network.

[0119] The parameter prediction error intuitively reflects the error between the predicted combustion parameters of the combustion parameter prediction network and the actual combustion parameters, and is the core index for measuring the prediction accuracy of the network, denoted as E₃. For example, when predicting the combustion efficiency, the difference between the network prediction value and the actual measured true combustion efficiency is the parameter prediction error.

[0120] The third error influence coefficient of the parameter prediction error is the coefficient that assigns the weight of the parameter prediction error in the comprehensive error calculation, denoted as γ. Since the parameter prediction error is directly related to the prediction accuracy of the network, usually γ takes the largest value among the three influence coefficients, that is, γ > β > α.

[0121] The computer system determines the comprehensive error through the following formula .

[0122] By executing step 5211, the computer system can comprehensively consider various error factors and their weights, and accurately calculate the comprehensive error. This comprehensive error fully reflects the performance of the combustion parameter prediction network in different aspects, provides an accurate quantitative basis for subsequent calibration of the a-th fusion component set and the second inference layer in the network, helps to gradually optimize the network performance, and improve its prediction accuracy of the combustion parameters of the ADN-based liquid propellant.

[0123] In one implementation, step 522, according to the comprehensive error, for the a-th fusion component set and the second inference layer in the combustion parameter prediction network, includes:

[0124] Step 5221: Perform error backpropagation according to the comprehensive error to iterate the network parameters of the a-th fusion component set and the second inference layer;

[0125] The method further includes:

[0126] Step 5222: After completing the error backpropagation iteration of the network parameters of the a-th fusion component set and the second inference layer according to the comprehensive error, perform error backpropagation according to the second probability distribution error to iterate the network parameters of the second inference layer.

[0127] In step 5221, perform error backpropagation according to the comprehensive error to iterate the network parameters of the a-th fusion component set and the second inference layer. After obtaining the comprehensive error, the computer system will use the backpropagation algorithm to reverse the error from the output layer to the previous layers of the network, so as to calculate the gradients of the error with respect to the network parameters (such as weights and biases) of the a-th fusion component set and the second inference layer.

[0128] Taking a simplified neural network structure as an example, assume that the a-th fusion component set in the combustion parameter prediction network contains a fully connected layer with 3 input neurons and 2 output neurons, the connection weight matrix is W, and the bias vector is b. The second inference layer is also a fully connected layer, receiving the output of the a-th fusion component set, with 2 input neurons and 1 output neuron, the weight matrix is V, and the bias is c.

[0129] When the computer system inputs the data into the network, after a series of calculations, the prediction result is obtained, and the comprehensive error E is obtained by comparing with the true value total . According to the backpropagation algorithm, the error will backpropagate from the output layer. First, calculate the gradient of the error with respect to the output of the second inference layer ( is the output of the second inference layer). Then, use the chain rule to calculate the gradients of the error with respect to the weight V and bias c of the second inference layer. For example, for the weight (i represents the input neuron index, j represents the output neuron index), its gradient For the bias c, the gradient 。

[0130] Next, the error continues to backpropagate to the a-th set of fusion components. Calculate the gradient of the error with respect to the output of the a-th set of fusion components (where h is the output of the a-th set of fusion components), and then calculate the gradients with respect to the weights W and bias b using the chain rule. For example, the gradient of the weight (where k represents the input neuron index and l represents the output neuron index) is , and the gradient of the bias b is 。

[0131] After calculating these gradients, the computer system updates the network parameters according to the gradient descent method. For the weight V of the second inference layer, the update formula is ; the update formula for the bias c is 。For the weight W of the a-th set of fusion components, the update formula is ; the update formula for the bias b is , where is the learning rate.

[0132] Through error backpropagation and parameter iteration, the network parameters of the a-th set of fusion components and the second inference layer are continuously adjusted, so that the comprehensive error gradually decreases and the prediction performance of the network is gradually improved.

[0133] In step 5222, after completing the error backpropagation iteration of the network parameters of the a-th set of fusion components and the second inference layer based on the comprehensive error, error backpropagation is performed according to the second probability distribution error to iterate the network parameters of the second inference layer. This step is to further finely tune the second inference layer, aiming to optimize the consistency and stability of the internal inference process of the network. After completing the parameter update based on the comprehensive error, the computer system considers the second probability distribution error separately. Recall that the second probability distribution error is the probability distribution gap between the predicted combustion parameters of the first inference layer and the second inference layer.

[0134] The computer system starts again from the second probability distribution error and uses the backpropagation algorithm to calculate the gradient of this error with respect to the network parameters of the second inference layer. Through this additional error backpropagation and parameter iteration, the computer system specifically optimizes the second inference layer for the second probability distribution error. This helps to further improve the stability of the prediction results of the second inference layer and its consistency with the first inference layer, thereby enhancing the reliability of the internal inference of the entire combustion parameter prediction network.

[0135] Steps 5221 and S5222 are key operations for the computer system to calibrate the combustion parameter prediction network. Through comprehensive parameter iteration based on the comprehensive error and fine calibration of the second inference layer based on the second probability distribution error, the prediction performance of the network is continuously improved, enabling more accurate prediction of the combustion parameters of ADN-based liquid propellants, providing more reliable results for subsequent combustion simulation and analysis, and promoting the development of related research and applications.

[0136] In one implementation, the method further includes:

[0137] Step 523: After the calibration of the a-th fusion component set is completed, determine the performance metric of the combustion parameter prediction network on the validation dataset;

[0138] Step 524: When the network performance metric meets the fusion requirements, based on the combustion parameter prediction network, fuse the component sets at the (a + 1)-th level in the X single-parameter prediction networks.

[0139] In step 523, after the calibration of the a-th fusion component set is completed, determine the performance metric of the combustion parameter prediction network on the validation dataset (such as accuracy, recall, F1-score, etc.). When the computer system completes the calibration of the a-th fusion component set, a quantitative metric is needed to evaluate the performance of the current combustion parameter prediction network, which requires performance measurement on the validation dataset.

[0140] The validation dataset is a set of pre-prepared data that is different from the training data but has similar characteristics. Its role is to evaluate the generalization ability of the network without affecting network training. Taking the prediction of the temperature distribution in the combustion chamber as an example, after the computer system completes the calibration of the a-th fusion component set, the data related to the combustion conditions in the validation dataset is input into the combustion parameter prediction network. After calculation, the network outputs the prediction results of the temperature distribution in the combustion chamber.

[0141] For the evaluation of the prediction results, the computer system uses multiple performance metric indicators. Accuracy is a common indicator that measures the proportion of correctly predicted samples in the total number of samples. Recall focuses on the proportion of samples that are actually positive and are correctly predicted as positive. The F1-score is the harmonic mean of accuracy and recall, which combines the information of both and more comprehensively reflects the performance of the network.

[0142] In addition to the above metrics, the computer system may also use other evaluation metrics, such as the mean squared error (MSE), to measure the error degree between the predicted temperature and the true temperature. Through these different evaluation metrics, the computer system can comprehensively understand the performance of the combustion parameter prediction network on the validation dataset from multiple perspectives.

[0143] In step 524, when the network effect metric meets the fusion requirements, based on the combustion parameter prediction network, the component sets at the (a + 1)-th level in the X single-parameter prediction networks are fused. When the effect metric of the combustion parameter prediction network calculated by the computer system on the validation dataset meets the preset fusion requirements, it will enter the next network construction operation.

[0144] The fusion requirements are a standard set according to the actual needs and expectations for network performance. For example, it is set that the F1 score needs to reach above 0.75, or the mean squared error needs to be lower than a certain specific value, etc. When the various metrics of the combustion parameter prediction network meet these requirements, it indicates that the current network state is good and it has the conditions for further fusion.

[0145] Taking X = 3 single-parameter prediction networks (predicting temperature, pressure, and component concentration respectively) as an example, after the combustion parameter prediction network has been tuned based on the a-th fusion component set and the effect metric meets the requirements, the computer system starts to fuse the component sets at the (a + 1)-th level in these 3 single-parameter prediction networks.

[0146] For the single-parameter network for temperature prediction, the component set at the (a + 1)-th level will perform further feature extraction and transformation on the data processed by the previous levels. Suppose this level is a convolutional layer, and it will slide the convolutional kernel over the data to extract higher-level features related to temperature. The component sets at the (a + 1)-th level of the single-parameter networks for pressure prediction and component concentration prediction will perform similar operations, extracting features related to pressure and component concentration respectively.

[0147] The computer system uses a suitable fusion method to fuse the component sets at the (a + 1)-th level of these 3 single-parameter prediction networks. One method is concatenation. For example, the output of the component set at the (a + 1)-th level of the temperature prediction network is a feature vector with dimension m, the output of this level of the pressure prediction network is a feature vector with dimension n, and the output of this level of the component concentration prediction network is a feature vector with dimension p. The computer system concatenates these vectors in sequence to form a new feature vector with dimension m + n + p as the fused result.

[0148] During the fusion process, the computer system needs to ensure that the component sets output by different single-parameter prediction networks are compatible in terms of data type, dimension, etc. If they are not compatible, some data conversion operations may be required, such as data normalization, dimension adjustment, etc. After the fusion is completed, a new set of (a + 1)-th fusion components is formed. This set integrates the information of multiple single-parameter prediction networks at the (a + 1)-th level, providing a richer feature representation for the combustion parameter prediction network. Then, based on this new set of fusion components, the computer system continues to build and tune the combustion parameter prediction network, repeating the previous training and evaluation processes to continuously improve the performance of the network.

[0149] In one implementation, the method provided by the embodiments of the present invention further includes:

[0150] Step 525: Determine the effectiveness metric of the single-parameter prediction network on the validation dataset;

[0151] Step 526: When the metric difference between the effectiveness metric of the combustion parameter prediction network and the effectiveness metric of the single-parameter prediction network is less than a preset metric difference, determine that the network effectiveness metric meets the fusion requirements.

[0152] In step 525, determine the effectiveness metric of the single-parameter prediction network on the validation dataset. During the process of tuning the combustion parameter prediction network, the computer system needs to clarify the performance of the single-parameter prediction network on the validation dataset, which serves as an important basis for comparison and evaluation. For each single-parameter prediction network, the computer system inputs the data in the validation dataset related to the combustion parameter predicted by the network. Taking the single-parameter prediction network for predicting the pressure change in the combustion chamber as an example, the validation dataset contains relevant data under various combustion conditions, such as the flow rate of the propellant, injection angle, initial state of the combustion chamber, etc. The computer system inputs this data into the single-parameter network for pressure prediction. After a series of calculations and processes by the network, it outputs the prediction result of the pressure change in the combustion chamber.

[0153] The computer system uses multiple metrics to measure the effectiveness of the single-parameter prediction network. Similar to the combustion parameter prediction network, the metrics include accuracy, recall rate, F1 score, etc.

[0154] For the single-parameter prediction network for predicting the temperature distribution in the combustion chamber and the single-parameter prediction network for predicting the component concentration, the computer system will also use the same method for evaluation, and calculate various effectiveness metric indicators on their respective validation data subsets.

[0155] In step 526, when the metric difference between the metric of the combustion parameter prediction network and the metric of the single-parameter prediction network is less than the preset metric difference, it is determined that the network metric meets the fusion requirement. After the computer system calculates the metric indexes of the combustion parameter prediction network and the single-parameter prediction network on the validation data set respectively, a comparative analysis will be carried out on the two.

[0156] The preset metric difference is a standard value set according to actual requirements and expectations for network performance. This value is used to judge whether the combustion parameter prediction network is sufficiently superior to the single-parameter prediction network in terms of performance, so as to determine whether it meets the requirements for further fusion. For example, the preset metric difference is set to 0.1.

[0157] The computer system calculates the difference between the combustion parameter prediction network and the single-parameter prediction network on the same evaluation index. When the metric differences of all relevant evaluation indexes calculated by the computer system are less than the preset metric difference, it can be determined that the network metric meets the fusion requirement.

[0158] By accurately determining the metric of the single-parameter prediction network on the validation data set through step 525, and judging whether the network metric meets the fusion requirement by comparing the metric differences in step 526, the computer system can comprehensively evaluate the performance of the combustion parameter prediction network, ensuring that when further fusion operations are carried out, the network can achieve a good balance between performance improvement and structural optimization, thus laying a solid foundation for more accurately predicting the combustion parameters of ADN-based liquid propellants and promoting the continuous optimization and improvement of combustion simulation analysis methods.

[0159] In one implementation, the execution data of the component set at the (a + 1)-th level is the output data of the component set at the a-th level; the method further includes:

[0160] Step 527: When the network metric does not meet the fusion requirement, abandon the fusion of the component set at the a-th level in the X single-parameter prediction networks, and stop the component set fusion.

[0161] When the computer system completes the comparison between the metric of the combustion parameter prediction network and the metric of the single-parameter prediction network, if it is found that the metric difference does not meet the preset metric difference, that is, when the network metric does not meet the fusion requirement, step 527 will be executed. For example, assume that when evaluating the network related to the prediction of the pressure in the combustion chamber, the preset metric difference is set to 0.1. After calculation, the accuracy rate of the combustion parameter prediction network in pressure prediction is 0.7, while the accuracy rate of the single-parameter network for pressure prediction is 0.8, and the metric difference between the two is 0.1. The metric differences of other evaluation indexes such as recall rate and F1 score also do not meet the requirements. This indicates that the current combustion parameter prediction network is not significantly superior to the single-parameter prediction network in terms of pressure prediction, and the overall network metric does not meet the fusion requirement.

[0162] In this case, the computer system abandons the fusion operation on the set of components at the a-th level in X single-parameter prediction networks. Continuing with the pressure prediction as an example, originally planned to fuse the set of components at the a-th level in X single-parameter prediction networks (assuming X = 3, namely temperature, pressure, and component concentration prediction networks), due to the network effect metric not meeting the requirements, the computer system stops this fusion behavior. This means that the set of components at the a-th level of the pressure prediction single-parameter network will not be concatenated or otherwise fused with the sets of components at the a-th level of the other two single-parameter prediction networks, avoiding the risk of network performance deterioration caused by unreasonable fusion.

[0163] By executing step 527, the computer system stops losses in a timely manner when the network effect is not ideal, ensuring that the calibration of the combustion parameter prediction network proceeds in the direction of improving performance. This helps save computing resources, avoid wasting time and computing power due to unnecessary fusion operations, and at the same time ensure the stability and rationality of the network structure, providing a basis for possible subsequent calibration strategy adjustments to better achieve accurate prediction of ADN-based liquid propellant combustion parameters.

[0164] In one implementation, the output layer of the combustion parameter prediction network includes the prediction output layers of X single-parameter prediction networks; the method further includes:

[0165] Step 60: After the calibration of the set of fusion components in the combustion parameter prediction network is completed, fine-tune the parameters of each prediction output layer in the combustion parameter prediction network according to the specific parameter prediction errors corresponding to each parameter prediction branch.

[0166] When the calibration of the set of fusion components in the combustion parameter prediction network is completed, it means that the overall network architecture and most of the key parts have been optimized. However, in order to further improve the prediction accuracy of the network for different combustion parameters, it is necessary to fine-tune the parameters of each prediction output layer.

[0167] Each parameter prediction branch corresponds to a specific combustion parameter prediction task, such as the temperature distribution prediction branch, the pressure change prediction branch, etc. The specific parameter prediction error refers to the error between the prediction result of each parameter prediction branch and the true combustion parameter. The computer system calculates this error for each parameter prediction branch respectively.

[0168] Taking the branch for predicting temperature distribution as an example, the computer system inputs a set of test data containing various combustion conditions into the combustion parameter prediction network. After a series of operations, the branch for predicting temperature distribution outputs the predicted temperature distribution results. These predicted results are compared with the true temperature distribution obtained by actual measurement, and the error is calculated. For example, the mean square error (MSE) is used for calculation. For the branch for predicting pressure change, the corresponding test data is also input, and the error between its predicted result and the true pressure change is calculated.

[0169] Based on these targeted parameter prediction errors, the computer system fine-tunes the parameters of each prediction output layer. The parameters of the prediction output layer mainly include weights and biases, etc.

[0170] In one implementation scheme, the output layer of the combustion parameter prediction network includes the prediction output layers of X single-parameter prediction networks; Step 40, setting X second inference layers for the a-th fusion component set, including:

[0171] Step 41: When the component set at the a-th level is not the last component set before the output layer of the parameter prediction branch, set X second inference layers for the a-th fusion component set;

[0172] The method provided by the embodiments of the present invention further includes:

[0173] Step 42: When the component set at the a-th level is the last component set before the output layer of the parameter prediction branch, use the true combustion parameters corresponding to the combustion condition training data of the ADN-based liquid propellant sample as the prior label, and determine the predicted combustion parameters of the single-parameter prediction network as the prior label of the predicted combustion parameters of the combustion parameter prediction network, and calibrate the combustion parameter prediction network.

[0174] In Step 41, when the component set at the a-th level is not the last component set before the output layer of the parameter prediction branch, set X second inference layers for the a-th fusion component set. During the construction process of the combustion parameter prediction network, the hierarchical structure is gradually built, and each level of component set plays a specific role. When the computer system determines that the component set at the a-th level is not the last component set before the output layer of the parameter prediction branch, this step will be executed.

[0175] Taking a simplified combustion parameter prediction network as an example, assume that the network has multiple levels for predicting three combustion parameters, namely the temperature distribution, pressure change, and component concentration in the combustion chamber, that is, X = 3. At a certain stage of the network, the component set at the a-th level has completed the feature extraction and preliminary processing of the input data, but there is still a certain hierarchical interval from the output layer.

[0176] For the temperature parameter prediction branch, the computer system sets up a second inference layer dedicated to temperature prediction for the a-th set of fusion components. This layer receives the output of the a-th set of fusion components and further analyzes and infers temperature-related features. For example, if the output of the a-th set of fusion components is a feature vector containing various combustion-related information, the second inference layer for temperature prediction will perform specific calculations on this vector. Suppose the feature vector is , the second inference layer may perform matrix multiplication (where is the weight matrix for temperature prediction of this second inference layer, and is the intermediate result related to temperature after preliminary calculations), and then through a non-linear transformation (such as the ReLU function: f(x)=max(0,x)), further extract and strengthen temperature-related features to prepare for accurately predicting the temperature distribution later.

[0177] For the pressure parameter prediction branch and the component concentration parameter prediction branch, the computer system will similarly set up corresponding second inference layers for the a-th set of fusion components respectively. The second inference layer for pressure prediction performs operations with the output of the a-th set of fusion components through a different weight matrix and through the corresponding non-linear transformation to process pressure-related information; the second inference layer for component concentration prediction uses the weight matrix to perform operations such as to focus on the analysis of component concentration-related features.

[0178] In step 42, when the set of components at the a-th level is the last set of components before the output layer of the parameter prediction branch, the true combustion parameters corresponding to the combustion condition training data of the ADN-based liquid propellant sample are used as prior labels, and the predicted combustion parameters of the single-parameter prediction network are determined as the prior labels for the predicted combustion parameters of the combustion parameter prediction network to calibrate the combustion parameter prediction network. When the set of components at the a-th level is at the last position before the output layer of the parameter prediction branch, the situation is different. At this time, the computer system uses specific prior labels to calibrate the combustion parameter prediction network.

[0179] Still taking the network with three parameter prediction branches mentioned above as an example, when the set of components at the a-th level is at the last position before the output layer of the temperature parameter prediction branch, the computer system uses the true temperature distribution data of the ADN-based liquid propellant sample under various combustion conditions as prior labels. These true data are obtained through precise experimental measurements and reflect the temperature situation in the actual combustion process. At the same time, the computer system also uses the predicted combustion parameters of the temperature single-parameter prediction network for the same combustion conditions as prior labels.

[0180] ​For the temperature parameter prediction branch of the combustion parameter prediction network, the computer system calibrates according to these prior markers. Assume that the predicted value of the temperature distribution under a certain combustion condition by the combustion parameter prediction network is , while the true temperature distribution is , and the predicted value of the temperature single-parameter prediction network is . The computer system calculates the error between the predicted value and the true value. For example, the mean square error (MSE) formula (n is the number of measurement positions), and the difference metric between the predicted value and the predicted value of the single-parameter prediction network (such as KL divergence: , where are the probability distributions of the predicted temperature values by the combustion parameter prediction network and the single-parameter prediction network respectively).

[0181] Based on these errors and difference metrics, the computer system adjusts the parameters related to the temperature parameter prediction branch in the combustion parameter prediction network. For the weights w and biases b in the network, they are adjusted in a manner similar to the gradient descent method. By continuously adjusting these parameters, the temperature prediction result of the combustion parameter prediction network gradually approaches the true value and the predicted value of the single-parameter prediction network, improving the accuracy and stability of the network.

[0182] For the pressure parameter prediction branch and the component concentration parameter prediction branch, the computer system will also calculate the errors and difference metrics and adjust the corresponding network parameters in the above manner, using their respective true combustion parameters and the predicted combustion parameters of the single-parameter prediction network as prior markers.

[0183] In one implementation, in step 20, calibrating the first inference layer corresponding to each component set in the single-parameter prediction network includes:

[0184] Step 21: Set the first inference layer corresponding to each component set in the single-parameter prediction network;

[0185] Step 22: Determine the inference error of the first inference layer based on the predicted combustion parameter of the first inference layer corresponding to the a-th component set in the single-parameter prediction network and the true combustion parameter;

[0186] Step 23: Calibrate the first inference layer based on the inference error.

[0187] In step 21, a corresponding first inference layer is set for each component set in the single-parameter prediction network. When the computer system constructs the single-parameter prediction network, according to the network structure and the requirements of the prediction task, a dedicated first inference layer is equipped for each component set. The component set can be understood as a combination of neurons at a certain level in the network. They jointly process the input data and extract features valuable for predicting combustion parameters. Taking a single-parameter prediction network for predicting the temperature distribution in a combustion chamber as an example, this network contains multiple levels. At a certain level, the component set receives the output data from the previous level, which may contain various information related to combustion, such as the composition of the propellant, the injection speed, the geometry of the combustion chamber, etc. A first inference layer is set for this component set. The role of this first inference layer is to further analyze and reason based on the features output by the component set to obtain a prediction result related to the temperature distribution.

[0188] In step 22, based on the predicted combustion parameters of the first inference layer corresponding to the a-th component set in the single-parameter prediction network and the true combustion parameters, the inference error of the first inference layer is determined. After setting the first inference layer for the component set, to evaluate the prediction effect of the first inference layer, the inference error needs to be calculated. The calculation method of the aforementioned mean square error can be referred to, and it will not be elaborated here.

[0189] In step 23, the first inference layer is calibrated according to the inference error. After calculating the inference error of the first inference layer, the first inference layer is adjusted according to these errors to improve its prediction accuracy. Taking the first inference layer of the temperature prediction single-parameter network as an example, assuming that the calculated mean square error is large, it indicates that there is a large deviation between the prediction result of the first inference layer and the true value, and it needs to be calibrated. The gradient descent method is used to adjust the parameters of the first inference layer, such as weights and biases.

[0190] For the weight W, assume its adjustment formula is , where is the learning rate, which controls the step size of weight adjustment. The learning rate is an important hyperparameter. If it is set too large, the weight update speed is too fast, which may cause the network not to converge; if it is set too small, the network training speed will be very slow. For example, assume that a certain element of the current weight is 0.5, and the calculated value at this element is 0.2, and the learning rate =0.01, then the updated weight value at this element is 0.5 - 0.01×0.2 = 0.498.

[0191] For the bias b, the adjustment formula is . For example, the current bias =0.3, = 0.1, the learning rate = 0.01, then the updated bias = 0.3 - 0.01 × 0.1 = 0.299.

[0192] By repeating this process multiple times, the input data under different combustion conditions is fed into the network, the inference error is calculated, and then the weights and biases are adjusted according to the error. After each adjustment, the inference error gradually decreases, and the prediction accuracy of the first inference layer continuously improves.

[0193] The embodiment of the present invention provides a computer system, as Figure 3 shown, the computer system 100 includes: a processor 101 and a memory 103. Among them, the processor 101 and the memory 103 are connected, such as connected through a bus 102. Optionally, the computer system 100 may further include a transceiver 104. It should be noted that in actual applications, the transceiver 104 is not limited to one, and the structure of the computer system 100 does not constitute a limitation to the embodiment of the present invention.

[0194] The embodiment of the present invention provides a computer system. The computer system in the embodiment of the present invention includes: one or more processors; a memory; one or more computer programs, where one or more computer programs are stored in the memory and configured to be executed by one or more processors. When the one or more programs are executed by the processor, the above-mentioned combustion simulation and analysis method based on ADN-based liquid propellant is implemented.

Claims

1. A combustion simulation and analysis method based on an ADN-based liquid propellant, characterized in that The method includes: In response to a simulation analysis instruction, obtaining combustion condition data of a target ADN-based liquid propellant; Loading the combustion condition data of the target ADN-based liquid propellant into a pre-tuned combustion parameter prediction network, where the combustion parameter prediction network is a joint learning network including multiple different branch parameter prediction networks, and one branch parameter prediction network is used to predict one combustion parameter; Outputting at least one combustion parameter through the combustion parameter prediction network; Performing visualization processing based on the at least one combustion parameter to obtain a simulation result of the target ADN-based liquid propellant; Among them, the combustion parameter prediction network is tuned through the following steps: Obtaining X single-parameter prediction networks, where different single-parameter prediction networks are used to predict different combustion parameters, and the component sets at the same level in different single-parameter prediction networks have the same network architecture, and X is a positive integer greater than 1; Tuning the first inference layer corresponding to the component set in the single-parameter prediction network, where the first inference layer is used to obtain a predicted combustion parameter based on the output of the component set; Fusing the component sets at the a-th level in the X single-parameter prediction networks to obtain a combustion parameter prediction network, where the combustion parameter prediction network corresponds to X parameter prediction branches, and the combustion parameter prediction network includes the first fusion component set to the a-th fusion component set obtained by fusion, and a is a positive integer greater than 0; Setting X second inference layers for the a-th fusion component set, where different second inference layers are used to obtain predicted combustion parameters corresponding to different parameter prediction branches based on the output of the fusion component set; Using the true combustion parameter corresponding to the combustion condition training data of the ADN-based liquid propellant sample as a prior label, using the predicted combustion parameter of the first inference layer as the prior label of the predicted combustion parameter of the second inference layer, and using the predicted combustion parameter of the single-parameter prediction network as the prior label of the predicted combustion parameter of the combustion parameter prediction network to tune the combustion parameter prediction network; the architecture of the a-th fusion component set is the same as that of the component set at the a-th level in the single-parameter prediction network, and the initial network parameters of the a-th fusion component set are the average values of the network parameters of the component sets at the a-th level in the X single-parameter prediction networks.

2. The method according to claim 1, wherein The step of using the true combustion parameter corresponding to the combustion condition training data of the ADN-based liquid propellant sample as a prior label, using the predicted combustion parameter of the first inference layer as the prior label of the predicted combustion parameter of the second inference layer, and using the predicted combustion parameter of the single-parameter prediction network as the prior label of the predicted combustion parameter of the combustion parameter prediction network to tune the combustion parameter prediction network includes: Determine the parameter prediction error, the first probability distribution error, and the second probability distribution error. The parameter prediction error represents the error between the predicted combustion parameter of the combustion parameter prediction network and the true combustion parameter. The first probability distribution error represents the probability distribution gap between the predicted combustion parameters of the combustion parameter prediction network and the single-parameter prediction network. The second probability distribution error represents the probability distribution gap between the predicted combustion parameters of the first inference layer and the second inference layer. Calibrate the a-th fusion component set and the second inference layer in the combustion parameter prediction network according to the parameter prediction error, the first probability distribution error, and the second probability distribution error.

3. The method according to claim 2, wherein The determination of the parameter prediction error, the first probability distribution error, and the second probability distribution error includes: Determine the parameter prediction error according to the predicted combustion parameters of X parameter prediction branches by the combustion parameter prediction network and the true combustion parameters corresponding to the X parameter prediction branches. For the same parameter prediction branch, determine the first predicted combustion parameter of the parameter prediction branch by the combustion parameter prediction network and the second predicted combustion parameter of the parameter prediction branch by the single-parameter prediction network corresponding to the parameter prediction branch. Determine the first probability distribution error according to the probability distributions of the first predicted combustion parameter and the second predicted combustion parameter. For the same parameter prediction branch, determine the third predicted combustion parameter of the second inference layer corresponding to the parameter prediction branch and the fourth predicted combustion parameter of the first inference layer corresponding to the parameter prediction branch. Determine the second probability distribution error according to the probability distributions of the third predicted combustion parameter and the fourth predicted combustion parameter.

4. The method according to claim 2, wherein The calibration of the a-th fusion component set and the second inference layer in the combustion parameter prediction network according to the parameter prediction error, the first probability distribution error, and the second probability distribution error includes: Determine the comprehensive error according to the parameter prediction error, the first probability distribution error, and the second probability distribution error. Calibrate the a-th fusion component set and the second inference layer in the combustion parameter prediction network according to the comprehensive error.

5. The method according to claim 4, characterized in that The determination of the comprehensive error according to the parameter prediction error, the first probability distribution error, and the second probability distribution error includes: Determine the comprehensive error according to the first probability distribution error, the first error influence coefficient of the first probability distribution error, the second probability distribution error, the second error influence coefficient of the second probability distribution error, the parameter prediction error, and the third error influence coefficient of the parameter prediction error. Wherein, the third error influence coefficient is greater than the second error influence coefficient, and the second error influence coefficient is greater than the first error influence coefficient. The calibration of the a-th fusion component set and the second inference layer in the combustion parameter prediction network according to the comprehensive error includes: Perform error backpropagation according to the comprehensive error to iterate the network parameters of the a-th fusion component set and the second inference layer. The method further includes: After the error backpropagation iteration of the network parameters of the a-th fusion component set and the second inference layer is completed according to the comprehensive error, error backpropagation is performed according to the second probability distribution error to iterate the network parameters of the second inference layer; The method further includes: After the calibration of the a-th fusion component set is completed, determine the combustion parameter prediction network effect metric of the combustion parameter prediction network on the validation data set; When the network effect metric meets the fusion requirements, based on the combustion parameter prediction network, fuse the component sets at the a+1-th level in the X single-parameter prediction networks.

6. The method according to claim 5, wherein The method further includes: Determine the single-parameter prediction network effect metric of the single-parameter prediction network on the validation data set; When the metric difference between the combustion parameter prediction network effect metric and the single-parameter prediction network effect metric is less than the preset metric difference, determine that the network effect metric meets the fusion requirements; The execution data of the component set at the a+1-th level is the output data of the component set at the a-th level; The method further includes: When the network effect metric does not meet the fusion requirements, abandon the fusion of the component sets at the a-th level in the X single-parameter prediction networks and stop the component set fusion.

7. The method according to claim 2, wherein The output layer of the combustion parameter prediction network includes the prediction output layers of the X single-parameter prediction networks; The method further includes: After the calibration of the fusion component set in the combustion parameter prediction network is completed, fine-tune the parameters of each prediction output layer in the combustion parameter prediction network according to the targeted parameter prediction errors corresponding to each parameter prediction branch.

8. The method according to any one of claims 1 to 7, characterized in that, The output layer of the combustion parameter prediction network includes the prediction output layers of the X single-parameter prediction networks; Setting X second inference layers for the a-th fusion component set includes: When the component set at the a-th level is not the last component set before the output layer of the parameter prediction branch, set X second inference layers for the a-th fusion component set; The method further includes: When the component set at the a-th level is the last component set before the output layer of the parameter prediction branch, use the true combustion parameters corresponding to the combustion condition training data of the ADN-based liquid propellant sample as the prior label, and determine the predicted combustion parameters of the single-parameter prediction network as the prior label of the predicted combustion parameters of the combustion parameter prediction network, and calibrate the combustion parameter prediction network; The calibration of the first inference layers corresponding to the component sets in the single-parameter prediction network includes: Set the corresponding first inference layers for the component sets in the single-parameter prediction network; Determine the inference error of the first inference layer according to the predicted combustion parameters of the first inference layer corresponding to the a-th component set in the single-parameter prediction network and the true combustion parameters; Calibrate the first inference layer according to the inference error.

9. A computer system, characterized in that, Includes: One or more processors; A memory; One or more computer programs; Wherein the one or more computer programs are stored in the memory and configured to be executed by the one or more processors, and when the one or more computer programs are executed by the processor, the method according to any one of claims 1 to 8 is implemented.

Images