Performance parameter determination method for engine regenerative cooling simulation

Abstract

The invention provides a performance parameter determination method for regenerative cooling simulation of an engine, and relates to the technical field of rocket engine simulation, and the method comprises the steps: obtaining a training data set which comprises a simulation data set and an experiment data set, carrying out the iterative training of a Bayesian neural network model according to the simulation data set and the experiment data set, and obtaining a performance parameter of the simulation data set; obtaining posterior distribution of the Bayesian neural network model; based on the posterior distribution, the performance parameters and the confidence score and the error range of each performance parameter can be quickly and accurately obtained according to the design parameters of the engine model, the simulation calculation time of the engine model is effectively shortened, and the simulation efficiency is improved. Moreover, according to the method provided by the invention, the posterior distribution is determined on the basis of the simulation data set obtained by simulation calculation and the experiment data set obtained by actual experiment, so that the alignment of the simulation data and the experiment data can be effectively realized, and the accuracy of the prediction result is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of rocket engine simulation, and in particular to a method for determining performance parameters for engine regenerative cooling simulation. Background Art

[0002] In recent years, with the rapid development of closed-cycle liquid rocket technology, the requirements for accurate calculation of temperature distribution in the thrust chamber of liquid rocket engines, optimization of cooling effects, and structural strength assessment have been continuously improved.

[0003] In the related art, numerical calculations are usually performed using Fluent algorithm software. However, this simulation calculation method exposes many problems in the early design stage. For example, simulating only a simplified engine model with a geometric structure of 1 / 100 usually requires a grid distribution of tens of millions, resulting in calculation time taking dozens of hours. This high computational workload and long calculation mode seriously limits the efficiency of the early design stage. To circumvent these problems, technicians often rely on personal experience to develop preliminary design plans, and then adjust the details of the plan to reduce the amount of simulation calculations. However, this approach based on personal experience is difficult to balance accuracy and efficiency, especially in complex design tasks, where the trade-off between efficiency and accuracy is even more difficult, and there is a problem of low computational efficiency.

[0004] To meet the demand for rapid iteration in the early design phase, researchers have recently attempted to incorporate machine learning techniques into the liquid rocket engine design process. For example, neural network models have been used to construct rapid prediction models. These models can significantly improve the efficiency of design evaluation in the early stages. However, despite some breakthroughs, this approach still has significant drawbacks. Current neural network models typically rely on simulation data for training, which inevitably carries model errors and deviates significantly from actual physical experimental data. This deviation manifests as insufficient accuracy in model predictions, making it difficult to truly reflect the actual situation under complex operating conditions. Furthermore, traditional neural network models are mostly "black box" models, unable to explain the internal logic of the prediction results and lacking clear quantification of the uncertainty of the results, which poses potential risks in design decisions. Furthermore, traditional neural network models are typically trained with fixed parameters, making them difficult to flexibly adjust to new experimental data, thus limiting their applicability in design iterations.

[0005] In addition to the field of liquid rocket engine thrust chamber design, similar problems are also widely present in other engineering design fields. For example, in the aerodynamic shape design of hypersonic vehicles, in order to evaluate the heat flow distribution and aerodynamic performance of the aircraft surface, traditional CFD simulations also face the problem of high computational cost and long time consumption. Existing accelerated prediction methods based on machine learning are also limited when applied to the airspace field due to insufficient model accuracy, difficulties in integrating experimental data, and the lack of uncertainty assessment. In addition, experimental data, as a key factor in improving model reliability, is often used inefficiently in existing technologies. Although some studies have attempted to combine simulation data and experimental data, a systematic integration mechanism has not yet been formed, which further restricts the improvement of model performance.

[0006] In summary, relevant technologies still have limitations in many aspects, including the lack of efficient and rapid prediction capabilities, insufficient integration and utilization of experimental data and simulation data, lack of reliability and uncertainty in prediction results, and flexibility in parameter adjustment. Summary of the Invention

[0007] An embodiment of the present invention provides a method for determining performance parameters for engine regenerative cooling simulation. The method can be based on a pre-trained neural network model and can quickly and accurately obtain performance parameters and the confidence score and error range of each performance parameter according to the design parameters of the engine model. It can enable technicians to quickly evaluate the risks of the engine through the confidence score and error range of each performance parameter, effectively reducing the simulation calculation time of the engine model and improving simulation efficiency. In addition, the method provided by the present invention can achieve assimilation or alignment of simulation data and experimental data, and the established Bayesian neural network model can simultaneously refer to the knowledge carried by the simulation data and experimental data to effectively provide the uncertainty of the prediction results. The Bayesian neural network model provided by the present invention can not only quantify the confidence of the prediction results, but also reveal the possible error range, helping decision makers to better understand and evaluate the reliability and risk of the model. In subsequent research, the Bayesian neural network model provided by the present invention can be embedded in the simulator to replace part of the calculation work, thereby significantly reducing the simulation calculation time while improving the flexibility and accuracy of the prediction.

[0008] To achieve the above objectives, the embodiments of the present invention adopt the following technical solutions:

[0009] In a first aspect, a method for determining performance parameters for engine regenerative cooling simulation is provided, the method comprising: obtaining a training data set, the training data set comprising a simulation data set and an experimental data set, the simulation data comprising design parameters of multiple engine models and performance parameters of each engine model obtained according to simulation calculations, the experimental data set comprising design parameters of multiple engine models and performance parameters of each engine model obtained according to simulation experiments, the design parameters comprising one or more of cooling channel geometry parameters, material parameters and flow rate, and the performance parameters comprising one or more of pressure drop, wall temperature, pressure and velocity; iteratively training a Bayesian neural network model according to the simulation data set to obtain a prior distribution of the Bayesian neural network model; updating the prior distribution according to the experimental data set to obtain a posterior distribution of the Bayesian neural network model; obtaining data to be tested, the data to be tested comprising the design parameters of a target engine model; and determining a predicted value and a confidence interval of each performance parameter of the target engine model according to the data to be tested based on the posterior distribution of the Bayesian neural network model.

[0010] In a possible implementation of the first aspect, before iteratively training the Bayesian neural network model based on the simulation data set to obtain the prior distribution of the Bayesian neural network model, the above method also includes: performing a preprocessing operation on the training data set, the preprocessing operation including an outlier deletion operation, a missing value filling operation and a normalization operation, and the outlier value is a value that is not within a preset threshold range.

[0011] In a possible implementation of the first aspect, the prior distribution is updated according to the experimental data set to obtain the posterior distribution of the Bayesian neural network model, including: constructing a variational distribution and an evidence lower bound function; based on the evidence lower bound function, iteratively updating the variational distribution according to the experimental data set; when the value of the evidence lower bound function is greater than a preset threshold, determining the variational distribution as the posterior distribution of the Bayesian neural network model.

[0012] In a possible implementation of the first aspect, based on the evidence lower bound function, the variational distribution is iteratively updated according to the experimental data set, including: determining the entropy corresponding to each latent variable of the variational distribution according to the experimental data set; determining the weight coefficient corresponding to each latent variable according to the entropy corresponding to each latent variable, wherein the weight coefficient corresponding to each latent variable is negatively correlated with the entropy; based on the evidence lower bound function, the variational distribution is iteratively updated according to the weight coefficient corresponding to each latent variable and the prior distribution.

[0013] In a possible implementation of the first aspect, the entropy H(θ i ) is determined by:

[0014]

[0015] Among them, σ i is the standard deviation of the i-th latent variable.

[0016] The weight coefficient λ of the i-th latent variable i The formula for determining is:

[0017] λ i =1 / (1+H(θ i )).

[0018] In a possible implementation of the first aspect, the evidence lower bound function L is:

[0019]

[0020] Among them, E q(θ) [logp(D | θ)] is the log-likelihood expectation term; D is the experimental data set, θ is the latent variable; q φ (θ i ) is the variational distribution; KL(q φ (θ i )||p(θ i )) is the KL divergence value of the i-th latent variable, p(θ i ) is the prior distribution; K is the number of latent variables; θ i is the i-th latent variable; λ i is the weight coefficient of the i-th latent variable.

[0021] The beneficial effects of the present invention are as follows: the method provided by the present invention first trains the prior distribution of the Bayesian neural network model based on the simulation data set obtained by simulation calculation, and then updates the prior distribution based on the experimental data set obtained by actual experiment to obtain the posterior distribution, and then can determine the predicted value and confidence interval of each performance parameter of the target engine model according to the posterior distribution of the Bayesian neural network model, which can enable technicians to quickly evaluate the risks of the engine through the confidence score and error range of each performance parameter, effectively reducing the simulation calculation time of the engine model and improving the simulation efficiency. In addition, the method provided by the present invention can realize the assimilation or alignment of simulation data and experimental data, and the established Bayesian neural network model can effectively provide the uncertainty of the prediction results. The Bayesian neural network model provided by the present invention can not only quantify the confidence of the prediction results, but also reveal the possible error range, helping decision makers to better understand and evaluate the reliability and risk of the model. In subsequent research, the model can be embedded in the simulator to replace part of the calculation work, thereby significantly reducing the simulation calculation time while improving the flexibility and accuracy of the prediction. On the other hand, the present invention provides a variational inference method for determining the posterior distribution, by assigning weight coefficients to different latent variables in the process of updating the posterior distribution of the Bayesian neural network model. Since the weight coefficient is determined based on the entropy of each latent variable, it is possible to effectively improve the accuracy of the solution of the latent variable, thereby improving the accuracy and reliability of the model. The method provided by the present invention further enhances the adaptability to system complexity and changes by modeling the uncertainty of the latent variables, and improves the performance of the model in practical applications. Therefore, the posterior distribution of the Bayesian neural network model provided by the present invention can improve the prediction accuracy and robustness of performance parameters.

[0022] In a second aspect, the present invention provides a performance parameter determination system for engine regenerative cooling simulation, the system comprising: a data acquisition module for acquiring a training data set, the training data set comprising a simulation data set and an experimental data set, the simulation data comprising design parameters of a plurality of engine models and performance parameters of each engine model obtained according to simulation calculations, the experimental data set comprising design parameters of a plurality of engine models and performance parameters of each engine model obtained according to simulation experiments, the design parameters comprising one or more of cooling channel geometry parameters, material parameters and flow rate, the performance parameters comprising one or more of pressure drop, wall temperature, pressure and velocity; a priori determination module for iteratively training a Bayesian neural network model according to the simulation data set to obtain a priori distribution of the Bayesian neural network model; a posterior determination module for updating the priori distribution according to the experimental data set to obtain a posterior distribution of the Bayesian neural network model; the data acquisition module for also acquiring data to be tested, the data to be tested comprising the design parameters of the target engine model; a parameter determination module for determining the predicted value and confidence interval of each performance parameter of the target engine model according to the data to be tested based on the posterior distribution of the Bayesian neural network model.

[0023] In a possible implementation of the second aspect, the posterior determination module is specifically used to: construct a variational distribution and an evidence lower bound function; based on the evidence lower bound function, iteratively update the variational distribution according to the experimental data set; when the value of the evidence lower bound function is greater than a preset threshold, determine the variational distribution as the posterior distribution of the Bayesian neural network model.

[0024] In a possible implementation of the second aspect, the posterior determination module is specifically configured to: determine the entropy corresponding to each latent variable of the variational distribution based on the experimental data set; determine a weight coefficient corresponding to each latent variable based on the entropy corresponding to each latent variable, wherein the weight coefficient corresponding to each latent variable is negatively correlated with the entropy; and iteratively update the variational distribution based on the weight coefficient corresponding to each latent variable and the prior distribution based on the evidence lower bound function;

[0025] The entropy H(θ corresponding to the i-th hidden variable i ) is determined by:

[0026]

[0027] Among them, σ i is the standard deviation of the i-th latent variable;

[0028] The weight coefficient λ of the i-th latent variable i The formula for determining is:

[0029] λ i =1 / (1+H(θ i )).

[0030] The evidence lower bound function L is:

[0031]

[0032] Among them, E q(θ) [logp(D|θ)] is the log-likelihood expectation term; D is the experimental data set, θ is the latent variable; q φ (θ i ) is the variational distribution; KL(q φ (θ i )||p(θ i )) is the KL divergence value of the i-th latent variable, p(θ i ) is the prior distribution; K is the number of latent variables; θ i is the i-th latent variable; λ i is the weight coefficient of the i-th latent variable.

[0033] In a third aspect, an electronic device is provided, comprising a memory and one or more processors; the memory is coupled to the processor; wherein computer program code is stored in the memory, and the computer program code comprises computer instructions, and when the computer instructions are executed by the processor, the electronic device executes a method as in any implementation of the first aspect.

[0034] In a fourth aspect, a computer-readable storage medium is provided, comprising computer instructions. When the computer instructions are executed on an electronic device, the electronic device executes the method in any implementation of the first aspect.

[0035] According to a fifth aspect, a computer program product is provided. When the computer program product is run on a computer, the computer is caused to execute the method in any implementation of the first aspect.

[0036] It can be understood that the beneficial effects that can be achieved by the system of the second aspect, the electronic device of the third aspect, the computer-readable storage medium of the fourth aspect, and the computer program product of the fifth aspect provided above can be referred to the beneficial effects in the first aspect and any possible design method thereof, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 A schematic structural diagram of an electronic device provided by an embodiment of the present invention;

[0038] Figure 2 A flow chart of a method for determining performance parameters for engine regenerative cooling simulation provided by an embodiment of the present invention;

[0039] Figure 3A flowchart of another method for determining performance parameters for engine regenerative cooling simulation provided by an embodiment of the present invention;

[0040] Figure 4 A flowchart of another method for determining performance parameters for engine regenerative cooling simulation provided by an embodiment of the present invention;

[0041] Figure 5 A schematic diagram of changes in the weight coefficients of latent variables during a training process according to an embodiment of the present invention;

[0042] Figure 6 A structural diagram of a determination system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0043] The technical solutions in the embodiments of the present invention will be described below with reference to the accompanying drawings in the embodiments of the present invention. In the description of the present invention, unless otherwise specified, " / " indicates that the objects associated before and after are in an "or" relationship. For example, A / B can represent A or B. The "or" in the present invention is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A or B can represent: A exists alone, A and B exist at the same time, and B exists alone. A and B can be singular or plural. In addition, in the description of the present invention, unless otherwise specified, "multiple" refers to two or more than two. "At least one of the following items" or similar expressions refers to any combination of these items, including any combination of single items or plural items.

[0044] In addition, to facilitate a clear description of the technical solutions of the embodiments of the present invention, in the embodiments of the present invention, the words "first" and "second" are used to distinguish between identical or similar items with substantially the same functions and effects. Those skilled in the art will understand that the words "first" and "second" do not limit the quantity or execution order, and the words "first" and "second" do not necessarily mean different.

[0045] In the embodiments of the present invention, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of the present invention should not be construed as superior or more advantageous than other embodiments or designs. Rather, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner to facilitate understanding.

[0046] In recent years, with the rapid development of closed-cycle liquid rocket technology, the requirements for accurate calculation of temperature distribution in the thrust chamber of liquid rocket engines, optimization of cooling effects, and structural strength assessment have been continuously improved.

[0047] In the related art, numerical calculations are usually performed using Fluent algorithm software. However, this simulation calculation method exposes many problems in the early design stage. For example, simulating only a simplified engine model with a geometric structure of 1 / 100 usually requires a grid distribution of tens of millions, resulting in calculation time taking dozens of hours. This high computational workload and long calculation mode seriously limits the efficiency of the early design stage. To circumvent these problems, technicians often rely on personal experience to develop preliminary design plans, and then adjust the details of the plan to reduce the amount of simulation calculations. However, this approach based on personal experience is difficult to balance accuracy and efficiency, especially in complex design tasks, where the trade-off between efficiency and accuracy is even more difficult, and there is a problem of low computational efficiency.

[0048] To meet the demand for rapid iteration in the early design phase, researchers have recently attempted to incorporate machine learning techniques into the liquid rocket engine design process. For example, neural network models have been used to construct rapid prediction models. These models can significantly improve the efficiency of design evaluation in the early stages. However, despite some breakthroughs, this approach still has significant drawbacks. Current neural network models typically rely on simulation data for training, which inevitably carries model errors and deviates significantly from actual physical experimental data. This deviation manifests as insufficient accuracy in model predictions, making it difficult to truly reflect the actual situation under complex operating conditions. Furthermore, traditional neural network models are mostly "black box" models, unable to explain the internal logic of the prediction results and lacking clear quantification of the uncertainty of the results, which poses potential risks in design decisions. Furthermore, traditional neural network models are typically trained with fixed parameters, making them difficult to flexibly adjust to new experimental data, thus limiting their applicability in design iterations.

[0049] In addition to the field of liquid rocket engine thrust chamber design, similar problems are also widely present in other engineering design fields. For example, in the aerodynamic shape design of hypersonic vehicles, in order to evaluate the heat flow distribution and aerodynamic performance of the aircraft surface, traditional CFD simulations also face the problem of high computational cost and long time consumption. Existing accelerated prediction methods based on machine learning are also limited when applied to the airspace field due to insufficient model accuracy, difficulties in integrating experimental data, and the lack of uncertainty assessment. In addition, experimental data, as a key factor in improving model reliability, is often used inefficiently in existing technologies. Although some studies have attempted to combine simulation data and experimental data, a systematic integration mechanism has not yet been formed, which further restricts the improvement of model performance.

[0050] In summary, the problems existing in the related art are as follows: First, there is a lack of effective assimilation of simulation and experimental data: Existing neural network models typically rely solely on large amounts of simulation data for training. However, simulation data often carries model errors and deviates significantly from real physical experimental data, resulting in insufficient reliability of model predictions. Second, existing neural network models cannot dynamically update model parameters: Existing neural network-based models are often "black box models" with fixed parameters. After training, they are difficult to dynamically adjust based on new experimental data, limiting the model's ability to continuously optimize during the actual design process. Third, existing neural network models lack prediction uncertainty: Traditional methods such as neural networks and support vector machines cannot effectively quantify the uncertainty of prediction results, which can easily lead to risks in design decisions, especially when the prediction results deviate significantly from the real data. Finally, neural network models in the related art do not effectively utilize experimental data. Although experimental data is of great value in improving model reliability, existing methods lack a systematic approach to combining simulation and experimental data, making it difficult to effectively integrate the two.

[0051] In summary, relevant technologies still have limitations in many aspects, including the lack of efficient and rapid prediction capabilities, insufficient integration and utilization of experimental data and simulation data, lack of reliability and uncertainty in prediction results, and flexibility in parameter adjustment.

[0052] In view of this, an embodiment of the present invention provides a performance parameter determination method for engine regenerative cooling simulation, the method comprising: obtaining a training data set, the training data set comprising a simulation data set and an experimental data set, the simulation data comprising design parameters of multiple engine models and performance parameters of each engine model obtained according to simulation calculations, the experimental data set comprising design parameters of multiple engine models and performance parameters of each engine model obtained according to simulation experiments, the design parameters comprising one or more of cooling channel geometry parameters, material parameters and flow rate, and the performance parameters comprising one or more of pressure drop, wall temperature, pressure and velocity; iteratively training a Bayesian neural network model according to the simulation data set to obtain a prior distribution of the Bayesian neural network model; updating the prior distribution according to the experimental data set to obtain a posterior distribution of the Bayesian neural network model; obtaining data to be tested, the data to be tested comprising the design parameters of the target engine model; and determining, based on the posterior distribution of the Bayesian neural network model, a predicted value and a confidence interval of each performance parameter of the target engine model according to the data to be tested.

[0053] The method provided by the present invention first trains the prior distribution of the Bayesian neural network model based on the simulation data set obtained by simulation calculation, and then updates the prior distribution based on the experimental data set obtained by actual experiment to obtain the posterior distribution. Then, the predicted value and confidence interval of each performance parameter of the target engine model can be determined based on the posterior distribution of the Bayesian neural network model, which enables technicians to quickly evaluate the risks of the engine through the confidence score and error range of each performance parameter, effectively reducing the simulation calculation time of the engine model and improving simulation efficiency. In addition, the method provided by the present invention can realize the assimilation or alignment of simulation data and experimental data, and the established Bayesian neural network model can effectively provide the uncertainty of the prediction results. The Bayesian neural network model provided by the present invention can not only quantify the confidence of the prediction results, but also reveal the possible error range, helping decision makers to better understand and evaluate the reliability and risk of the model. In subsequent research, the model can be embedded in the simulator to replace part of the calculation work, thereby significantly reducing the simulation calculation time while improving the flexibility and accuracy of the prediction. On the other hand, the present invention provides a variational inference method for determining the posterior distribution, by assigning weight coefficients to different latent variables in the process of updating the posterior distribution of the Bayesian neural network model. Since the weight coefficient is determined based on the entropy of each latent variable, it is possible to effectively improve the accuracy of the solution of the latent variable, thereby improving the accuracy and reliability of the model. The method provided by the present invention further enhances the adaptability to system complexity and changes by modeling the uncertainty of the latent variables, and improves the performance of the model in practical applications. Therefore, the posterior distribution of the Bayesian neural network model provided by the present invention can improve the prediction accuracy and robustness of performance parameters.

[0054] In some embodiments, a performance parameter determination method for engine regenerative cooling simulation provided by an embodiment of the present invention may be executed by a performance parameter determination system 100 for engine regenerative cooling simulation (hereinafter referred to as determination system 100 ).

[0055] As an example, the determination system 100 may be any electronic device 200 with data processing capabilities, such as a general-purpose computer, a personal computer, a laptop computer, a switch, or a tablet computer, and the specific implementation of the determination system 100 is not limited here.

[0056] Figure 1 The electronic device 200 includes a processor 210 , a memory 220 , and a communication interface 230 .

[0057] The processor 210 may include one or more processing cores. The processor 210 uses various interfaces and lines to connect various parts of the electronic device 200, and executes various functions of the electronic device 200 and processes data by running or executing instructions, programs, code sets or instruction sets stored in the memory 220, and calling data stored in the memory 220. Optionally, the processor 210 can be implemented in the form of at least one hardware of a central processing unit (CPU), a graphics processing unit (GPU), a digital signal processing (DSP), a field-programmable gate array (FPGA), and a programmable logic array (PLA).

[0058] The memory 220 may include a random access memory (RAl) or a read-only memory (ROL). Optionally, the memory 220 includes a non-transitory computer-readable storage medium (NMT). The memory 220 may be used to store instructions, programs, codes, code sets, or instruction sets. The memory 220 may include a program storage area. The program storage area may store instructions for implementing an operating system, instructions for implementing at least one function (such as a video acquisition function, a feature extraction function, and a process detection function), instructions for implementing the above-mentioned various method embodiments, etc.

[0059] The communication interface 230 is used to communicate with other devices, equipment or communication networks, such as data storage devices, image processing equipment or Ethernet, radio access network (RAN), wireless local area network (WLAN), etc.

[0060] In physical implementation, the aforementioned components (e.g., processor 210, memory 220, and communication interface 230) may be components within the same device (e.g., a laptop). Alternatively, at least two of the components may be provided within the same device, i.e., as different components within a single device, similar to the deployment of devices or components in a distributed system.

[0061] It should be understood that the structure illustrated in this embodiment does not constitute a specific limitation on the electronic device 200. In other embodiments of the present invention, the electronic device 200 may include more or fewer components than shown, or may combine or separate certain components, or arrange the components differently. The illustrated components may be implemented in hardware, software, or a combination of software and hardware.

[0062] The following describes a method for determining performance parameters for engine regenerative cooling simulation provided by an embodiment of the present invention in conjunction with the accompanying drawings.

[0063] Figure 2 A flow chart of a method for determining performance parameters for engine regenerative cooling simulation provided by an embodiment of the present invention. Optionally, the method may be Figure 1 The electronic device 200 shown is executed. The method may include the following steps:

[0064] S1. Obtain a training data set, which includes a simulation data set and an experimental data set.

[0065] Specifically, the simulation data includes the design parameters of multiple engine models and the performance parameters of each engine model obtained based on simulation calculations. The experimental data set includes the design parameters of multiple engine models and the performance parameters of each engine model obtained based on simulation experiments. The design parameters include one or more of the cooling channel geometry parameters, material parameters and flow rate. The performance parameters include one or more of the pressure drop, wall temperature, pressure and velocity.

[0066] S2. Iteratively train the Bayesian neural network model based on the simulation data set to obtain the prior distribution of the Bayesian neural network model.

[0067] Specifically, the Bayesian Neural Network (BNN) model combines Bayesian probabilistic methods with neural networks, aiming to provide uncertainty measures for the weights and outputs of neural networks. Unlike traditional deterministic neural networks, BNNs introduce probability distributions to describe the uncertainty of model parameters. This allows them to not only provide prediction results but also assess the confidence of these results. In other words, the Bayesian Neural Network model incorporates Bayesian inference into neural networks. It not only learns mapping relationships in the data but also models the uncertainty of network parameters. Traditional neural networks are trained with a fixed set of connection parameters (weights and biases) and then make deterministic predictions for new data. In contrast, the Bayesian Neural Network model treats these parameters as random variables, assigning each parameter a prior distribution that is updated to a corresponding posterior distribution through observational data. Consequently, the BNN output contains not only the predicted value but also its uncertainty information (such as confidence intervals or variance). The core principle of the Bayesian Neural Network model is to use Bayes' theorem to combine prior knowledge with observed data to obtain the posterior distribution of the parameters. This uncertainty is then transferred to the predicted results to form a predictive distribution. This enables the model to make more robust judgments when faced with unknown or noisy data.

[0068] The prior distribution is a connection parameter between different layers included in the Bayesian neural network model, wherein the connection parameter includes a weight distribution and a bias distribution.

[0069] In some embodiments, the Bayesian neural network model includes multiple input layers, multiple hidden layers, and multiple output layers, and the prior distribution includes connection parameters between any two layers in the Bayesian neural network model that have a connection relationship. For example, the Bayesian neural network model includes input layer 1, input layer 2, hidden layer 1, and output layer 1, and the prior distribution includes a weight distribution and a bias distribution between input layer 1 and hidden layer 1, a weight distribution and a bias distribution between input layer 2 and hidden layer 1, and a weight distribution and a bias distribution between hidden layer 1 and output layer 1.

[0070] It should be understood that the Bayesian neural network model may include more or fewer input layers, hidden layers, and output layers than the above examples, and the embodiment of the present invention does not impose any particular limitation on this.

[0071] In a possible implementation, before the above S2, the method provided by the embodiment of the present invention further includes:

[0072] The training data set is preprocessed, including outlier removal, missing value filling, and normalization. Outliers are values that are not within a preset threshold range.

[0073] Specifically, the determination system 100 first filters outliers in the training data set, then supplements missing values in the training data set, and finally normalizes the training data set to normalize the values to a preset range to complete the preprocessing of the training data set.

[0074] S3. Update the prior distribution according to the experimental data set to obtain the posterior distribution of the Bayesian neural network model.

[0075] In some embodiments, see Figure 3 , the above S3 specifically includes:

[0076] S31. Construct variational distribution and evidence lower bound function.

[0077] The evidence lower bound function L is:

[0078]

[0079] Among them, E q(θ) [logp(D|θ)] is the log-likelihood expectation term; D is the experimental data set, θ is the latent variable; q φ (θ i ) is the variational distribution; KL(q φ (θ i )||p(θ i )) is the KL divergence value of the i-th latent variable, p(θ i ) is the prior distribution; K is the number of latent variables; θ i is the i-th latent variable; λ i is the weight coefficient of the i-th latent variable.

[0080] Specifically, the evidence lower bound function is used to measure the closeness between the variational distribution and the true posterior distribution, the log likelihood logp(D|θ) is used to measure the degree of match between the model prediction value and the true value carried in the experimental dataset, and the expectation E q(θ) The parameter used to make the distribution cover the region of interpretable data.

[0081] S32. Based on the evidence lower bound function, the variational distribution is iteratively updated according to the experimental dataset.

[0082] In one possible implementation, see Figure 4 The above S32 specifically includes:

[0083] S321. Determine the entropy corresponding to each latent variable of the variational distribution according to the experimental data set;

[0084] Specifically, the prior distribution includes the connection parameters between any two connected layers in the Bayesian neural network model. The connection parameters include the weight distribution and the bias distribution. The weight distribution includes the weight mean and weight standard deviation, and the bias distribution includes the bias mean and bias standard deviation. In other words, the latent variables are the distribution parameters of the weights and biases of each layer (such as the mean and standard deviation in a Gaussian distribution). Entropy is used to characterize the certainty of the latent variables.

[0085] In some embodiments, the entropy H(θ i ) is determined by:

[0086]

[0087] Among them, σ i is the standard deviation of the i-th latent variable.

[0088] S322. Determine a weight coefficient corresponding to each latent variable according to the entropy corresponding to each latent variable, wherein the weight coefficient corresponding to each latent variable is negatively correlated with the entropy;

[0089] Specifically, the weight coefficient λ of the i-th latent variable i The formula for determining is:

[0090] λ i =1 / (1+H(θ i )).

[0091] S323. Based on the evidence lower bound function, the variational distribution is iteratively updated according to the weight coefficient and prior distribution corresponding to each latent variable.

[0092] Specifically, the variational distribution is defined by all latent variables, and iterative updating of the variational distribution can be understood as adjusting the shape of the variational distribution (such as moving mean, scaled variance).

[0093] S33. When the value of the evidence lower bound function is greater than a preset threshold, the variational distribution is determined as the posterior distribution of the Bayesian neural network model.

[0094] Specifically, when the value of the evidence lower bound function is greater than a preset threshold, the error between the variational distribution and the true posterior distribution is less than a preset error interval, and thus the variational distribution is determined as the posterior distribution of the Bayesian neural network model.

[0095] S4. Acquire data to be tested, where the data to be tested includes design parameters of the target engine model.

[0096] S5. Based on the posterior distribution of the Bayesian neural network model, determine the predicted value and confidence interval of each performance parameter of the target engine model according to the data to be tested.

[0097] Exemplarily, when the performance parameters of the target engine model include wall temperature and pressure drop, the predicted value and confidence interval of the wall temperature are 314±4K, and the predicted value and confidence interval of the wall temperature are 0.2±0.01 MPa.

[0098] As can be seen from the above S1-S5, the method provided by the present invention first trains the prior distribution of the Bayesian neural network model based on the simulation data set obtained by simulation calculation, and then updates the prior distribution based on the experimental data set obtained by actual experiment to obtain the posterior distribution, and then can determine the predicted value and confidence interval of each performance parameter of the target engine model according to the posterior distribution of the Bayesian neural network model, which can enable technicians to quickly evaluate the risks of the engine through the confidence score and error range of each performance parameter, effectively reducing the simulation calculation time of the engine model and improving simulation efficiency. In addition, the method provided by the present invention can realize the assimilation or alignment of simulation data and experimental data, and the established Bayesian neural network model can effectively provide the uncertainty of the prediction results. The Bayesian neural network model provided by the present invention can not only quantify the confidence of the prediction results, but also reveal the possible error range, helping decision makers to better understand and evaluate the reliability and risk of the model. In subsequent research, the model can be embedded in the simulator to replace part of the calculation work, thereby significantly reducing the simulation calculation time while improving the flexibility and accuracy of the prediction.

[0099] In one possible implementation, the method provided by the embodiment of the present invention further includes:

[0100] The preset interface displays the change curve of the weight coefficient corresponding to each latent variable during the iterative training process.

[0101] For example, see Figure 5 , Figure 5 This diagram illustrates the changes in the weight coefficients of latent variables during training, as shown in an embodiment of the present invention. It includes two latent variables: parameter w1-loc and parameter w2-loc. The curve that gradually increases and remains relatively stable with the number of iterations represents the weight coefficient of parameter w1-loc for each iteration, while the curve that gradually decreases with the number of iterations represents the weight coefficient of parameter w2-loc for each iteration.

[0102] The method provided by the embodiment of the present invention can enable technicians to quickly and accurately determine the weight coefficients of different latent variables by displaying the change curves of the weight coefficients of different latent variables during the training process, and then

[0103] In order to facilitate understanding of this solution, the specific implementation process of the embodiment of the present invention is explained below through an example. First, a Bayesian neural network model is defined, and the weights and biases of the Bayesian neural network model are defined as random variables, and the initial prior distribution is a standard Gaussian distribution. Then, the Bayesian neural network model is iteratively trained through a simulation data set to obtain a trained prior distribution. It can also be understood that the initial prior distribution is optimized and updated according to the simulation data set to obtain an updated prior distribution, and this updated prior distribution incorporates the knowledge carried by the simulation data set. Then, an experimental data set is obtained, and the updated prior distribution is iteratively trained through the experimental data set to obtain a posterior distribution, wherein, in the process of iterative training of the updated prior distribution through the experimental data set, the objective function is the evidence lower bound function, and each hidden variable in the evidence lower bound function is dynamically assigned different weight coefficients based on entropy. Finally, the Bayesian neural network model can determine the performance parameters and confidence intervals corresponding to each data to be detected based on the posterior distribution.

[0104] That is, the method provided by the embodiment of the present invention completes the initial update of the prior distribution through the simulation data set, and then further updates the prior distribution based on the more realistic experimental data set obtained by the experiment to obtain the posterior distribution. In this way, the performance parameters of the target engine model determined by the posterior distribution can be close to the knowledge contained in the simulation data set and the experimental data set, thereby effectively improving the accuracy and robustness, meeting the requirements of accurate calculation of the temperature distribution of the thrust chamber of the liquid rocket engine, optimization of the cooling effect, and structural strength assessment. In other words, the embodiment of the present invention provides a variational inference method for determining the posterior distribution. By assigning weight coefficients to different latent variables in the process of updating the posterior distribution of the Bayesian neural network model, since the weight coefficients are determined based on the entropy of each latent variable, the accuracy of the latent variable solution can be effectively improved, thereby improving the accuracy and reliability of the model. The method provided by the present invention further enhances the adaptability to system complexity and changes by modeling the uncertainty of the latent variables, and improves the performance of the model in practical applications. Therefore, the posterior distribution of the Bayesian neural network model provided by the present invention can improve the prediction accuracy and robustness of performance parameters.

[0105] The above mainly introduces the solution of the embodiment of the present invention from the perspective of method. It can be understood that in order to realize the above functions, the determination system 100 includes at least one of the hardware structure and software modules corresponding to the execution of each function. It should be easy for those skilled in the art to realize that, in combination with the units and algorithm steps of each example described in the embodiment disclosed herein, the embodiment of the present invention can be implemented in the form of hardware or a combination of hardware and computer software. Whether a function is executed in the form of hardware or computer software driving hardware depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the embodiment of the present invention.

[0106] In an embodiment of the present invention, the determination system 100 may be divided into functional units according to the above-described method example. For example, the determination system 100 may be divided into functional units corresponding to respective functions, or two or more functions may be integrated into a single processing unit. The above-described integrated units may be implemented in the form of hardware or software functional units. It should be noted that the division of units in the embodiment of the present invention is illustrative and is merely a logical functional division. In actual implementation, other division methods may be used.

[0107] For example, Figure 6 The hardware structure diagram of a determination system provided by an embodiment of the present invention is shown. The determination system 100 includes: a data acquisition module 110 for acquiring a training data set, the training data set including a simulation data set and an experimental data set, the simulation data including design parameters of multiple engine models and performance parameters of each engine model obtained according to simulation calculations, the experimental data including design parameters of multiple engine models and performance parameters of each engine model obtained according to simulation experiments, the design parameters including one or more of cooling channel geometry parameters, material parameters, and flow rate, and the performance parameters including one or more of pressure drop, wall temperature, pressure, and velocity; a priori determination module 120 for iteratively training a Bayesian neural network model based on the simulation data set to obtain a priori distribution of the Bayesian neural network model; a posterior determination module 130 for updating the priori distribution based on the experimental data set to obtain a posterior distribution of the Bayesian neural network model; the data acquisition module 110 is further configured to acquire data to be tested, the data to be tested including design parameters of a target engine model; and a parameter determination module 140 for determining a predicted value and a confidence interval for each performance parameter of the target engine model based on the posterior distribution of the Bayesian neural network model.

[0108] Optionally, the posterior determination module 130 is specifically used to: construct a variational distribution and an evidence lower bound function; based on the evidence lower bound function, iteratively update the variational distribution according to the experimental data set; when the value of the evidence lower bound function is greater than a preset threshold, determine the variational distribution as the posterior distribution of the Bayesian neural network model.

[0109] Optionally, the posterior determination module 130 is specifically configured to: determine the entropy corresponding to each latent variable of the variational distribution based on the experimental data set; determine the weight coefficient corresponding to each latent variable based on the entropy corresponding to each latent variable, wherein the weight coefficient corresponding to each latent variable is negatively correlated with the entropy; and iteratively update the variational distribution based on the weight coefficient corresponding to each latent variable and the prior distribution based on the evidence lower bound function;

[0110] The entropy H(θ corresponding to the i-th hidden variable i ) is determined by:

[0111]

[0112] Among them, σ i is the standard deviation of the i-th latent variable;

[0113] The weight coefficient λ of the i-th latent variable i The formula for determining is:

[0114] λ i =1 / (1+H(θ i ));

[0115] The evidence lower bound function L is:

[0116]

[0117] Among them, E q(θ) [logp(D|θ)] is the log-likelihood expectation term; D is the experimental data set, θ is the latent variable; q φ (θ i ) is the variational distribution; KL(q φ (θ i )||p(θ i )) is the KL divergence value of the i-th latent variable, p(θ i ) is the prior distribution; K is the number of latent variables; θ i is the i-th latent variable; λ i is the weight coefficient of the i-th latent variable.

[0118] It should be understood that the specific description of the above optional methods can refer to the above method embodiments, which will not be repeated here. In addition, the explanation of any of the above-provided determination systems 100 and the description of the beneficial effects can refer to the above corresponding method embodiments, which will not be repeated here.

[0119] An embodiment of the present invention further provides a computer-readable storage medium storing at least one computer instruction, which is loaded and executed by a processor to implement the methods of each of the above embodiments. For explanations of the relevant contents and descriptions of the beneficial effects of any of the above-mentioned computer-readable storage media, reference can be made to the corresponding embodiments described above and will not be repeated here.

[0120] The embodiment of the present invention further provides a chip. The chip integrates a control circuit and one or more ports for implementing the functions of the above-mentioned determination system 100. Optionally, the functions supported by the chip can be referred to above and will not be repeated here.

[0121] Those skilled in the art will appreciate that all or part of the steps of the above-mentioned embodiments can be implemented by instructing the relevant hardware through a program, and the program can be stored in a computer-readable storage medium. The above-mentioned storage medium can be a read-only memory, a random access memory, etc. The above-mentioned processing unit or processor can be a central processing unit, a general-purpose processor, a specific circuit structure (application specific integrated circuit, ASIC), a microprocessor (digital signal processor, DSP), a field programmable gate array (field programmable gate array, FPGA) or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof.

[0122] An embodiment of the present invention further provides a computer program product comprising instructions that, when executed on a computer, cause the computer to perform any of the methods described in the above embodiments. The computer program product comprises one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions according to the embodiments of the present invention are fully or partially generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via a wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) method. The computer-readable storage medium may be any available medium accessible by a computer or a data storage device such as a server or data center that includes one or more available media. Available media may be magnetic media (e.g., floppy disks, hard disks, tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives).

[0123] It should be noted that the above-mentioned devices for storing computer instructions or computer programs provided in the embodiments of the present invention, such as but not limited to the above-mentioned memories, computer-readable storage media and communication chips, etc., all have non-transitory properties. Those skilled in the art should be aware that in one or more of the above examples, the functions described in the embodiments of the present invention can be implemented using hardware, software, firmware or any combination thereof. When implemented using software, these functions can be stored in a computer-readable storage medium or transmitted as one or more instructions or codes on a computer-readable storage medium. Computer-readable storage media include computer storage media and communication media, wherein the communication medium includes any medium that facilitates the transmission of a computer program from one place to another. The storage medium can be any available medium that can be accessed by a general-purpose or special-purpose computer.

[0124] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and are not to be construed as limitations on the present invention. A person skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. A method for determining performance parameters for engine regenerative cooling simulation, characterized in that: The method comprises: Acquiring a training data set, the training data set comprising a simulation data set and an experimental data set, the simulation data comprising design parameters of a plurality of engine models and performance parameters of each engine model obtained according to simulation calculations, the experimental data set comprising design parameters of a plurality of engine models and performance parameters of each engine model obtained according to simulation experiments, the design parameters comprising one or more of cooling channel geometry parameters, material parameters, and flow rate, and the performance parameters comprising one or more of pressure drop, wall temperature, pressure, and velocity; Iteratively training the Bayesian neural network model according to the simulation data set to obtain a prior distribution of the Bayesian neural network model; Updating the prior distribution according to the experimental data set to obtain the posterior distribution of the Bayesian neural network model; Acquiring data to be tested, wherein the data to be tested includes design parameters of a target engine model; Based on the posterior distribution of the Bayesian neural network model, a predicted value and a confidence interval of each performance parameter of the target engine model are determined according to the data to be tested.

2. The method according to claim 1, characterized in that Before iteratively training the Bayesian neural network model according to the simulation data set to obtain the prior distribution of the Bayesian neural network model, the method further includes: A preprocessing operation is performed on the training data set, wherein the preprocessing operation includes an outlier deletion operation, a missing value filling operation, and a normalization operation, wherein the outlier value is a value that is not within a preset threshold range.

3. The method according to claim 2, characterized in that The updating of the prior distribution according to the experimental data set to obtain the posterior distribution of the Bayesian neural network model includes: Construct variational distribution and evidence lower bound function; Based on the evidence lower bound function, iteratively updating the variational distribution according to the experimental data set; When the value of the evidence lower bound function is greater than a preset threshold, the variational distribution is determined as the posterior distribution of the Bayesian neural network model.

4. The method according to claim 3, characterized in that The iterative updating of the variational distribution based on the evidence lower bound function and the experimental data set includes: Determine the entropy corresponding to each latent variable of the variational distribution according to the experimental data set; The weight coefficient corresponding to each latent variable is determined according to the entropy corresponding to each latent variable, wherein the weight coefficient corresponding to each latent variable is negatively correlated with the entropy; Based on the evidence lower bound function, the variational distribution is iteratively updated according to the weight coefficient and prior distribution corresponding to each latent variable.

5. The method according to claim 4, characterized in that The entropy H(θ corresponding to the i-th hidden variable i ) is determined by: Among them, σ i is the standard deviation of the i-th latent variable; The weight coefficient λ of the i-th latent variable i The formula for determining is: l i =1 / (1+H(θi)).

6. The method according to claim 5, characterized in that The evidence lower bound function L is: Among them, E q(θ) [logp(D|θ)] is the log-likelihood expectation term; D is the experimental data set, θ is the latent variable; q φ (θ i ) is the variational distribution; KL(q φ (θ i )||p(θ i )) is the KL divergence value of the i-th latent variable, p(θ i ) is the prior distribution; K is the number of latent variables; θ i is the i-th latent variable; λ i is the weight coefficient of the i-th latent variable.

7. A performance parameter determination system for engine regenerative cooling simulation, characterized in that: The system comprises: a data acquisition module, configured to acquire a training data set, the training data set comprising a simulation data set and an experimental data set, the simulation data comprising design parameters of a plurality of engine models and performance parameters of each engine model obtained according to simulation calculations, the experimental data set comprising design parameters of a plurality of engine models and performance parameters of each engine model obtained according to simulation experiments, the design parameters comprising one or more of cooling channel geometry parameters, material parameters, and flow rate, and the performance parameters comprising one or more of pressure drop, wall temperature, pressure, and velocity; A priori determination module is used to iteratively train the Bayesian neural network model according to the simulation data set to obtain the prior distribution of the Bayesian neural network model; a posterior determination module, configured to update the prior distribution according to the experimental data set to obtain the posterior distribution of the Bayesian neural network model; The data acquisition module is further used to acquire data to be tested, wherein the data to be tested includes design parameters of the target engine model; A parameter determination module is used to determine the predicted value and confidence interval of each performance parameter of the target engine model based on the posterior distribution of the Bayesian neural network model and according to the data to be tested.

8. The system according to claim 7, characterized in that The a posteriori determination module is specifically used to: Construct variational distribution and evidence lower bound function; Based on the evidence lower bound function, iteratively updating the variational distribution according to the experimental data set; When the value of the evidence lower bound function is greater than a preset threshold, the variational distribution is determined as the posterior distribution of the Bayesian neural network model.

9. The system according to claim 8, characterized in that The a posteriori determination module is specifically used to: Determine the entropy corresponding to each latent variable of the variational distribution according to the experimental data set; The weight coefficient corresponding to each latent variable is determined according to the entropy corresponding to each latent variable, wherein the weight coefficient corresponding to each latent variable is negatively correlated with the entropy; Based on the evidence lower bound function, the variational distribution is iteratively updated according to the weight coefficient and prior distribution corresponding to each latent variable; The entropy H(θ corresponding to the i-th hidden variable i ) is determined by: Among them, σ i is the standard deviation of the i-th latent variable; The weight coefficient λ of the i-th latent variable i The formula for determining is: l i =1 / (1+H(θ i )); The evidence lower bound function L is: Among them, E q(θ) [logp(D|θ)] is the log-likelihood expectation term; D is the experimental data set, θ is the latent variable; q φ (θ i ) is the variational distribution; KL(q φ (θ i )||p(θ i )) is the KL divergence value of the i-th latent variable, p(θ i ) is the prior distribution; K is the number of latent variables; θ i is the i-th latent variable; λ i is the weight coefficient of the i-th latent variable.

10. An electronic device, characterized in that: include: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the instructions to implement the performance parameter determination method for engine regenerative cooling simulation according to any one of claims 1 to 6.

Images