Model determination method and device, storage medium and product
By using a neural network model-based screening and training method, a dataset that meets the constraint rules is generated, which solves the problems of high time cost and insufficient accuracy in aerodynamic calculation of aero-engine compressors, and achieves efficient and accurate aerodynamic calculation.
Patent Information
- Application Number
- CN202411103733.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-12
- Publication Date
- 2026-02-24
AI Technical Summary
The aerodynamic calculation model of the aero-engine compressor is time-consuming and costly, and the accuracy of the model is insufficient.
A dataset based on compressor experimental data was generated, and a neural network model was used for screening and training. Design rules were integrated to generate a dataset that satisfies the constraint rules, and the neural network model was trained for aerodynamic calculations.
It improves the efficiency and accuracy of compressor aerodynamic calculations, reduces calculation time, and enhances the consistency between the model and the actual design.
Smart Images
Figure CN121562331A_ABST
Abstract
Description
Technical Field
[0001] This disclosure relates to the field of aero-engine technology, and in particular to a method, apparatus, storage medium and product for determining a model. Background Technology
[0002] As the performance requirements for aero-engines increase, engineering design optimization problems become increasingly complex, necessitating numerical simulations to evaluate performance under different design parameters. Particularly in aero-engine compressor aerodynamic calculations, the solution space is large and complex; a single simulation can take minutes, hours, or even days to complete. Tasks such as design optimization, design space search, and sensitivity analysis require thousands or even millions of simulations, resulting in extremely high computational costs. Summary of the Invention
[0003] One of the technical problems this disclosure aims to solve is how to improve the accuracy of the models used for aerodynamic calculations of compressors.
[0004] According to some embodiments of this disclosure, a method for determining a model is provided, comprising: generating a dataset based on experimental data of a compressor; extracting constraint rules based on the design rules of the compressor; inputting the dataset and constraint rules into a neural network model to filter out data in the dataset that meets the constraint rules through the neural network model, wherein the neural network model is used to perform aerodynamic calculations of the compressor; and training the neural network model based on the data that meets the constraint rules to obtain a trained neural network model.
[0005] In some embodiments, extracting constraint rules based on compressor design rules includes: obtaining a document describing the design rules, wherein the design rules include aerodynamic design rules for the compressor and historical anomaly rules for the aerodynamic design of the compressor, and the aerodynamic design rules include design rules for the aerodynamic parameters of the compressor; and processing the design rules in the document into a standard format to generate constraint rules.
[0006] In some embodiments, processing the design rules in the document into a standard format to generate constraint rules includes: generating positive constraint rules based on the compressor's aerodynamic design rules in the standard format, and generating reverse constraint rules based on the historical anomaly rules in the standard format, wherein the positive constraint rules are used to indicate the normal aerodynamic design of the compressor, and the reverse constraint rules are used to indicate the abnormal aerodynamic design of the compressor.
[0007] In some embodiments, the constraint rules include a description of the compressor's design parameters and the setting of their ranges, and a description of the compressor's performance parameters and the setting of their ranges.
[0008] In some embodiments, the dataset includes data corresponding to multiple design parameters and multiple performance parameters of the compressor. Each data item in the dataset includes an input item and an output item, where the input item is the design parameter of the compressor and the output item is the performance parameter of the compressor.
[0009] In some embodiments, the inputs include at least one of inlet angle of attack, outlet lag angle, and metal angle of the compressor blades, and the outputs include at least one of pressure ratio, flow rate, efficiency, and surge margin.
[0010] In some embodiments, training a neural network model based on data that satisfies constraint rules includes: in each training round, inputting multiple data items from the dataset into the neural network model, comparing the output of the neural network model with the output items corresponding to the input items, determining the value of the loss function of the neural network model, and adjusting the parameters of the neural network model according to the value of the loss function until a preset convergence condition is reached.
[0011] In some embodiments, training a neural network model based on data that satisfies the constraint rules includes: dividing the data that satisfies the constraint rules into a training set and a test set; training the neural network model using the training set; and testing the trained neural network model using the test set.
[0012] In some embodiments, the determining method further includes: determining whether the trained neural network model meets the design requirements based on the accuracy of the trained neural network model on the test set.
[0013] In some embodiments, the determination method further includes: determining the neural network model and its complexity based on the number of design parameters included in the compressor design requirements.
[0014] In some embodiments, the determination method further includes: performing aerodynamic calculations and obtaining the values of compressor performance parameters by inputting the values of compressor design parameters into a trained neural network model.
[0015] According to some other embodiments of this disclosure, a model determination apparatus is provided, comprising: a generation module configured to generate a dataset based on experimental data of a compressor; an extraction module configured to extract constraint rules based on design rules of the compressor; a filtering module configured to input the dataset and constraint rules into a neural network model to filter out data in the dataset that satisfy the constraint rules through the neural network model, wherein the neural network model is used to perform aerodynamic calculations of the compressor; and a training module configured to train the neural network model based on the data that satisfy the constraint rules to obtain a trained neural network model.
[0016] According to further embodiments of this disclosure, a model determination apparatus is provided, comprising: a processor; and a memory coupled to the processor for storing instructions that, when executed by the processor, cause the processor to perform the model determination method as described above.
[0017] According to further embodiments of the present disclosure, a computer-readable storage medium is provided having a computer program stored thereon, wherein the program, when executed by a processor, implements the determination method as described above.
[0018] According to further embodiments of this disclosure, a computer program product is provided, including instructions that, when executed by a processor, cause the processor to perform the determination method as described above.
[0019] This disclosure uses a neural network model to filter experimental data from compressors to obtain a dataset that meets the compressor's design rules. The neural network model is then trained based on this dataset, enabling it to be used for compressor aerodynamic calculations. This not only improves the efficiency of compressor aerodynamic calculations but also enhances the accuracy of these calculations by incorporating the compressor's design rules into the neural network model.
[0020] Other features and advantages of this disclosure will become clear from the following detailed description of exemplary embodiments with reference to the accompanying drawings. Attached Figure Description
[0021] To more clearly illustrate the technical solutions in the embodiments of this disclosure or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0022] Figure 1 A flowchart illustrating a method for determining a model according to some embodiments of the present disclosure is shown.
[0023] Figure 2 A flowchart illustrating a method for determining a model according to other embodiments of this disclosure is shown.
[0024] Figure 3 A schematic diagram of the structure of a model determining device according to some embodiments of the present disclosure is shown.
[0025] Figure 4 A schematic diagram of the structure of a model determining device according to other embodiments of the present disclosure is shown.
[0026] Figure 5A schematic diagram of the structure of a model determining device according to some embodiments of the present disclosure is shown. Detailed Implementation
[0027] The technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit this disclosure or its application or use. All other embodiments obtained by those skilled in the art based on the embodiments of this disclosure without creative effort are within the scope of protection of this disclosure.
[0028] The air compressor is a crucial component of an aircraft engine. It compresses the air entering the engine, increasing its pressure and density, before sending it into the combustion chamber, thereby improving combustion efficiency and increasing the engine's output power.
[0029] To address the challenges and high costs associated with optimizing the aerodynamic design of compressors, this disclosure employs a surrogate model, specifically an artificial neural network (ANN) model, to simulate a high-precision compressor model. Artificial Neural Networks (ANNs), also known as Neural Networks (NNs) or Connection Models, are deep learning methods that mimic the behavioral characteristics of animal neural networks, employing distributed parallel information processing algorithms.
[0030] A surrogate model can achieve computational results that are very close to the original model, but with a smaller computational load. The surrogate model is built using a data-driven, bottom-up approach, that is, assuming that the internal precise processing of the original simulation process is unknown, the surrogate model is built by calculating the response (output) of the original model at a carefully selected finite number of points (inputs).
[0031] Figure 1 A flowchart illustrating a method for determining a model according to some embodiments of this disclosure is shown. Figure 1 As shown, the method of this embodiment includes steps S102 to S108.
[0032] In step S102, a dataset is generated based on the experimental data from the compressor.
[0033] In order to obtain a sufficiently accurate surrogate model, the number of sample points in the dataset cannot be too small, so it is necessary to collect sufficient experimental data of the compressor.
[0034] In some embodiments, the experimental data for the compressor can be numerical simulation data based on the compressor's historical aerodynamic model, or it can be based on actual aerodynamic test data of the compressor. Historical aerodynamic model refers to a highly accurate model of the compressor designed previously.
[0035] This approach not only allows for the collection of a sufficient number of sample points, but also enables the training of models based on historical aerodynamic shapes to incorporate information from historical designs. Consequently, the trained models can incorporate historical designs, thereby improving the efficiency of model training and the accuracy of the resulting models.
[0036] In some embodiments, the dataset includes data corresponding to multiple design parameters and multiple performance parameters of the compressor. Each data entry in the dataset includes an input item and an output item. The input item is the compressor's design parameters, and the output item is the compressor's performance parameters. The input item serves as the initial data for subsequent model training and as the training termination criterion.
[0037] In some embodiments, the dataset can be represented as a matrix consisting of parameters and their corresponding values. This facilitates the visualization of the dataset and allows for intervention when necessary.
[0038] In some embodiments, the inputs include at least one of inlet angle of attack, outlet lag angle, compressor blade metal angle, and blade pitch angle, and the outputs include at least one of pressure ratio, flow rate, efficiency, and surge margin.
[0039] In step S104, constraint rules are extracted based on the compressor design rules.
[0040] Compressor design rules refer to the parameters and ranges required to achieve compressor performance that meets target specifications. These design rules are often derived by summarizing historical design rules, which include both positive and negative rules. Positive rules are those that facilitate achieving the target performance, such as rules corresponding to successful historical designs. Negative rules, on the other hand, are those that hinder achieving the target performance, such as rules corresponding to failed historical designs. Compressor design rules typically include aerodynamic design requirements, aerodynamic design domains, aerodynamic analysis specifications, and aerodynamic calculation error evaluation criteria. Aerodynamic design requirements, for example, refer to the compressor's performance and stability requirements. The aerodynamic design domain refers to the range of compressor design parameters. The aerodynamic analysis specifications are the rules for analyzing the compressor's performance and stability. The aerodynamic calculation error evaluation criteria are the criteria for comparing the calculated values of the compressor's performance and stability parameters with the target values.
[0041] Because compressor design is quite complex, it generates a large number of historical design rules. By summarizing these historical rules to obtain design rules, and then extracting constraint rules based on these design rules, the complex historical design rules can be transformed into concise constraint rules. Designing compressors using constraint rules can reduce labor costs, facilitate batch design, and reduce the burden of continuous updates to constraint rules, thus making it easier to manage compressor design rules.
[0042] In some embodiments, extracting constraint rules based on compressor design rules includes: obtaining a document describing the design rules, wherein the design rules include aerodynamic design rules for the compressor and historical anomaly rules for the aerodynamic design of the compressor, and the aerodynamic design rules include design rules for the aerodynamic parameters of the compressor; and processing the design rules in the document into a standard format to generate constraint rules.
[0043] In some embodiments, the constraint rules include a description of the compressor's design parameters and the setting of their ranges, and a description of the compressor's performance parameters and the setting of their ranges.
[0044] In some embodiments, the standard format is a question-and-answer format, that is, the design rules are processed into a question-and-answer format. For example, a design rule regarding the inlet angle of attack is: What is the reasonable range of the inlet angle of attack? The reasonable range of the inlet angle of attack is {a, b}. Where a and b represent the lower limit and upper limit of the inlet angle of attack, respectively. Processing the design rules into a question-and-answer format is not only concise but also facilitates subsequent queries. That is, it is convenient not only for subsequent steps as input to the neural network model but also for manual queries, so that the design rules can be used as a compressor design library.
[0045] In some embodiments, processing the design rules in the document into a standard format to generate constraint rules includes: generating positive constraint rules based on the compressor's aerodynamic design rules in the standard format, and generating reverse constraint rules based on the historical anomaly rules in the standard format, wherein the positive constraint rules are used to indicate the normal aerodynamic design of the compressor, and the reverse constraint rules are used to indicate the abnormal aerodynamic design of the compressor.
[0046] By using positive and negative constraint rules, the compressor dataset can be constrained from two directions in subsequent steps, thereby quickly selecting a dataset that meets the requirements and improving the accuracy of the trained model.
[0047] In some embodiments, constraint rules are extracted from compressor design rules using machine learning models. For example, natural language processing models, neural network models, etc., can be used to extract constraint rules from compressor design rules.
[0048] In step S106, the dataset and constraint rules are input into the neural network model so that the neural network model can filter out the data in the dataset that meets the constraint rules. The neural network model is used to perform aerodynamic calculations for the compressor.
[0049] Neural network models include Deep Neural Networks (DNN), Convolutional Neural Networks (CNN), and Recurrent Neural Networks (RNN), which are characterized by their ability to handle complex simulations with large computational demands on nonlinear / nonconvex structures.
[0050] By using a neural network model to remove outliers that do not conform to the constraint rules, i.e., the aerodynamic response rules, the probability of introducing historical anomalous designs during model training is reduced.
[0051] It is understandable that the neural network model identifies outliers in the dataset after the neural network model has been trained accordingly. That is, it uses a neural network model capable of identifying abnormal data from the compressor to identify outliers in the dataset.
[0052] In some embodiments, the neural network model includes a preprocessing module, which can be pre-trained to identify constraint rules and remove outliers from the dataset. The neural network model also includes an execution module for performing aerodynamic calculations for the compressor. Therefore, the execution module of the neural network model can be trained using a dataset with outliers removed, enabling the neural network model to provide aerodynamic calculations for the compressor.
[0053] In some embodiments, a neural network model is used to confirm that the sample set includes data on the upper and lower boundaries of the compressor's aerodynamic performance, which improves the effectiveness of subsequent model use. For data in the sample set that do not include the upper and lower boundaries of aerodynamic performance, numerical simulation or experiments can be used to supplement the data.
[0054] By filtering the dataset using a neural network model and then training the model with data that meets the constraints, the neural network model incorporates information from the constraints. The model no longer relies solely on the dataset for training and testing to improve accuracy; it can also utilize the constraints as supplementary information. This is equivalent to connecting the neural network model to a database. This database can be used to integrate engineering experience and rules such as aerodynamic design requirements, analysis specifications, and error judgment criteria for engine compressors into the training process of the neural network model. This results in a higher consistency between the trained model and the original model or compressor, thus improving the accuracy of subsequent aerodynamic calculations using the neural network model.
[0055] In some embodiments, the determination method further includes: determining the neural network model and its complexity based on the number of design parameters included in the compressor design requirements.
[0056] In some embodiments, for single-stage compressor aerodynamic calculation models with fewer than 5 design parameters, simple artificial neural network algorithms such as linear neural networks are used to improve model training efficiency. For single-stage / multi-stage compressor aerodynamic calculation models with more than 5 design parameters, hierarchical artificial neural network algorithms such as deep neural networks are used to improve model training accuracy.
[0057] In step S108, the neural network model is trained based on the data that satisfies the constraint rules to obtain a trained neural network model.
[0058] In some embodiments, training a neural network model based on data that satisfies constraint rules includes: dividing the data into a training set and a test set; training the neural network model using the training set; and testing the trained neural network model using the test set. The data satisfying the constraint rules can be randomly classified; for example, most of the data can be used as the training set, and a small portion as the test set.
[0059] In some embodiments, training a neural network model based on data that satisfies constraint rules includes: in each training round, inputting multiple data items from the dataset into the neural network model, comparing the output of the neural network model with the output items corresponding to the input items, determining the value of the loss function of the neural network model, and adjusting the parameters of the neural network model according to the value of the loss function until a preset convergence condition is reached.
[0060] Loss functions can be set based on mean squared error, mean absolute error, cross-entropy loss, exponential loss, etc.
[0061] In some embodiments, the method for determining the model further includes determining whether the trained neural network model meets the design requirements based on the accuracy of the trained neural network model on the test set. For example, if the accuracy of the neural network model reaches 95%, the neural network model is considered to meet the design requirements.
[0062] In some embodiments, different target accuracy rates can be set for different performance parameters. For example, the target accuracy rates for efficiency, pressure ratio, and flow rate can be set to 95% or higher, while the target accuracy rate for surge margin can be set to 71.9% or higher.
[0063] The neural network model, as a proxy model for aerodynamic calculations of the compressor, is trained based on numerical simulation and experimental data. It can replace traditional numerical simulation and quickly obtain the aerodynamic calculation results of the compressor, serving as the model input for the optimization of the compressor and aero-engine as a whole.
[0064] In some embodiments, the method for determining the model further includes: inputting the values of the compressor's design parameters into the trained neural network model to perform aerodynamic calculations and obtain the values of the compressor's performance parameters. For example, by setting arbitrary inlet angle of attack, outlet lag angle, compressor blade metal angle, and blade pitch angle (the values of these parameters are all within the upper and lower limits set during the model training phase), rapid calculations of pressure ratio, flow rate, efficiency, and surge margin can be achieved.
[0065] This disclosure uses a neural network model to filter experimental data from compressors to obtain a dataset that meets the compressor's design rules. The neural network model is then trained based on this dataset, enabling it to be used for compressor aerodynamic calculations. This not only improves the efficiency of compressor aerodynamic calculations but also enhances the accuracy of these calculations by incorporating design rules into the neural network model.
[0066] Figure 2 A flowchart illustrating a method for determining a model according to other embodiments of this disclosure is shown. For example... Figure 2 As shown, the method of this embodiment includes steps S202 to S222.
[0067] In step S202, the design rules for the compressor are obtained.
[0068] In step S204, the aerodynamic shape of the compressor is obtained according to the compressor design rules.
[0069] In step S206, data simulation and / or aerodynamic testing are performed based on the aerodynamic model of the compressor to generate a dataset.
[0070] In step S208, the dataset and constraint rules are input into the neural network model so that the neural network model can filter the data in the dataset to remove outliers.
[0071] The extraction of constraint rules can be referred to the aforementioned embodiments, and will not be repeated here.
[0072] In some embodiments, the neural network model includes a preprocessing module and an execution module. The neural network model removes outliers from the dataset through the preprocessing module, while the execution module is used to perform aerodynamic calculations for the compressor.
[0073] In some embodiments, an optimization solver and loss function are selected based on an artificial neural network algorithm library to determine the execution module of the neural network, and the execution module of the neural network model is subsequently trained so that the neural network model can be used for aerodynamic calculations of the compressor.
[0074] In step S210, data extraction is performed on the dataset after outlier removal for use in training the subsequent neural network model. In some embodiments, data covering the upper and lower boundaries of the compressor's aerodynamic performance is extracted during data extraction.
[0075] Each data item in the dataset includes an input item and an output item. That is, the dataset includes input data and output data. The input data serves as the initial data for model iteration, and the output data serves as the termination criterion for model iteration.
[0076] In step S212, the dataset is divided into a training set and a test set.
[0077] In step S214, the neural network model is trained based on the training set, including steps S2141 to S2143.
[0078] In step S2141, the parameters are initialized. The initial values of the parameters are set according to the input data in the dataset.
[0079] In step S2142, the neural network model is trained.
[0080] In step S2143, it is determined whether the neural network meets the convergence condition. The convergence condition can be set based on the output data in the dataset, or the maximum number of iterations can be set. If yes, a surrogate model for compressor aerodynamic calculation is obtained. Otherwise, return to step S2141.
[0081] In step S216, a test data matrix is generated based on the test set. The test data includes input data and output data.
[0082] In step S218, the input data from the test set is input into the compressor aerodynamic calculation surrogate model, and the calculation results of the surrogate model are obtained.
[0083] In step S220, the calculation results of the proxy model are compared with the output data in the test set.
[0084] In step S222, the suitor model is deemed qualified based on the comparison results and the compressor aerodynamic calculation error evaluation criteria. For example, the compressor aerodynamic calculation error evaluation criteria might be a pass rate of at least 95%. If yes, training is complete. Otherwise, return to step S208.
[0085] Figure 3A schematic diagram of the structure of a model determining apparatus according to some embodiments of the present disclosure is shown. Figure 3 As shown, the device 30 in this embodiment includes modules 310-340.
[0086] The generation module 310 is configured to generate a dataset based on compressor-based experimental data;
[0087] Extraction module 320 is configured to extract constraint rules based on compressor design rules;
[0088] The filtering module 330 is configured to input the dataset and constraint rules into the neural network model so as to filter out the data in the dataset that meets the constraint rules through the neural network model. The neural network model is used to perform aerodynamic calculations for the compressor.
[0089] Training module 340 is configured to train a neural network model based on data that satisfies the constraint rules, so as to obtain a trained neural network model.
[0090] In some embodiments, the extraction module 320 is configured to obtain a document describing design rules, wherein the design rules include aerodynamic design rules for the compressor and historical anomaly rules for the aerodynamic design of the compressor, and the aerodynamic design rules include design rules for the aerodynamic parameters of the compressor; and process the design rules in the document into a standard format to generate constraint rules.
[0091] In some embodiments, the extraction module 320 is configured to generate positive constraint rules based on the compressor's aerodynamic design rules in a standard format, and to generate reverse constraint rules based on the historical anomaly rules in a standard format, wherein the positive constraint rules are used to indicate the normal aerodynamic design of the compressor, and the reverse constraint rules are used to indicate the abnormal aerodynamic design of the compressor.
[0092] In some embodiments, the constraint rules include a description of the compressor's design parameters and the setting of their ranges, and a description of the compressor's performance parameters and the setting of their ranges.
[0093] In some embodiments, the dataset includes data corresponding to multiple design parameters and multiple performance parameters of the compressor. Each data item in the dataset includes an input item and an output item, where the input item is the design parameter of the compressor and the output item is the performance parameter of the compressor.
[0094] In some embodiments, the inputs include at least one of inlet angle of attack, outlet lag angle, and metal angle of the compressor blades, and the outputs include at least one of pressure ratio, flow rate, efficiency, and surge margin.
[0095] In some embodiments, the training module 340 is configured to, in each round of training, input multiple data items from the dataset into the neural network model, compare the output of the neural network model with the output items corresponding to the input items, determine the value of the loss function of the neural network model, and adjust the parameters of the neural network model according to the value of the loss function until a preset convergence condition is reached.
[0096] In some embodiments, the training module 340 is configured to divide data that satisfies the constraint rules into a training set and a test set; train the neural network model using the training set; and test the trained neural network model using the test set.
[0097] In some embodiments, the determining device 30 is further configured to determine whether the trained neural network model is a neural network model that meets the design requirements based on the accuracy of the trained neural network model on the test set.
[0098] In some embodiments, the determining device 30 is further configured to determine the neural network model and its complexity based on the number of design parameters included in the compressor's design requirements.
[0099] In some embodiments, the determining device 30 is further configured to perform aerodynamic calculations and obtain the values of the compressor's performance parameters by inputting the values of the compressor's design parameters into the trained neural network model.
[0100] The model determination device disclosed herein uses a neural network model to filter experimental data of the compressor to obtain a dataset that meets the compressor's design rules. The neural network model is then trained based on this dataset, enabling it to be used for aerodynamic calculations of the compressor. This not only improves the efficiency of aerodynamic calculations but also enhances the accuracy of these calculations by incorporating design rules into the neural network model.
[0101] The model determination device in the embodiments of this disclosure can be implemented by various computing devices or computer systems, as described below. Figure 4 as well as Figure 5 Describe it.
[0102] Figure 4 A schematic diagram of the structure of a determining device for a model according to other embodiments of the present disclosure is shown. For example... Figure 4 As shown, the apparatus 40 of this embodiment includes a memory 410 and a processor 420 coupled to the memory 410. The processor 420 is configured to execute the model determination method in any of the embodiments of this disclosure based on instructions stored in the memory 410.
[0103] The memory 410 may include, for example, system memory, fixed non-volatile storage media, etc. The system memory may store, for example, the operating system, application programs, boot loader, database, and other programs.
[0104] Figure 5 A schematic diagram of the structure of a determining device according to some embodiments of the present disclosure is shown. Figure 5 As shown, the device 50 in this embodiment includes a memory 510 and a processor 520, which are similar to the memory 410 and processor 420, respectively. It may also include an input / output interface 530, a network interface 540, a storage interface 550, etc. These interfaces 530, 540, 550, and the memory 510 and processor 520 can be connected, for example, via a bus 560. The input / output interface 530 provides a connection interface for input / output devices such as a display, mouse, keyboard, and touchscreen. The network interface 540 provides a connection interface for various networked devices, such as connecting to a database server or cloud storage server. The storage interface 550 provides a connection interface for external storage devices such as SD cards and USB flash drives.
[0105] Embodiments of this disclosure also provide a computer-readable storage medium storing a computer program thereon, characterized in that the program, when executed by a processor, implements the method for determining any of the aforementioned models.
[0106] Embodiments of this disclosure also provide a computer program product, including instructions that, when executed by a processor, cause the processor to perform a determination method according to any of the foregoing models.
[0107] Those skilled in the art will understand that embodiments of this disclosure can be provided as methods, systems, or computer program products. Therefore, this disclosure can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this disclosure can take the form of a computer program product embodied on one or more computer-usable non-transitory storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0108] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0109] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0110] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0111] The above description is only a preferred embodiment of this disclosure and is not intended to limit this disclosure. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this disclosure should be included within the protection scope of this disclosure.
Claims
1. A method for determining a model, comprising: Datasets were generated based on experimental data from the compressor; Constraint rules are extracted based on the compressor's design rules; The dataset and the constraint rules are input into the neural network model to filter out the data in the dataset that satisfy the constraint rules. The neural network model is used to perform aerodynamic calculations for the compressor. Based on the data that satisfies the constraint rules, the neural network model is trained to obtain a trained neural network model.
2. The determination method according to claim 1, wherein, The constraint rules extracted based on the compressor design rules include: Obtain the document describing the design rules, wherein the design rules include the aerodynamic design rules of the compressor, the historical anomaly rules of the aerodynamic design of the compressor, and the aerodynamic design rules include the design rules of the aerodynamic parameters of the compressor; The design rules in the document are processed into a standard format to generate constraint rules.
3. The determination method according to claim 2, wherein, The step of processing the design rules in the document into a standard format to generate constraint rules includes: A positive constraint rule is generated based on the aerodynamic design rules of the compressor in the standard format, and a negative constraint rule is generated based on the historical anomaly rules in the standard format. The positive constraint rule is used to indicate the normal aerodynamic design of the compressor, and the negative constraint rule is used to indicate the abnormal aerodynamic design of the compressor.
4. The determination method according to claim 2, wherein, The constraint rules include the description of the design parameters of the compressor and the setting of their ranges, and the description of the performance parameters of the compressor and the setting of their ranges.
5. The determination method according to claim 1, wherein, The dataset includes data corresponding to multiple design parameters and multiple performance parameters of the compressor. Each data item in the dataset includes an input item and an output item. The input item is the design parameter of the compressor, and the output item is the performance parameter of the compressor.
6. The determining method according to claim 5, wherein, The input items include at least one of the following: inlet angle of attack, outlet lag angle, and metal angle of the compressor blades; the output items include at least one of the following: pressure ratio, flow rate, efficiency, and surge margin.
7. The determining method according to claim 5, wherein, Training the neural network model based on the data that satisfies the constraint rules includes: In each training round, the neural network model is input by multiple data items from the dataset, and the output of the neural network model is compared with the output corresponding to the input item to determine the value of the loss function of the neural network model. The parameters of the neural network model are then adjusted according to the value of the loss function until the preset convergence condition is reached.
8. The determination method according to claim 1, wherein, Training the neural network model based on the data that satisfies the constraint rules includes: The data that satisfies the constraint rules are divided into a training set and a test set; The neural network model is trained using the training set and tested using the test set.
9. The determining method according to claim 8, further comprising: The accuracy of the trained neural network model on the test set is used to determine whether the trained neural network model meets the design requirements.
10. The determining method according to claim 1, further comprising: The neural network model and its complexity are determined based on the number of design parameters included in the compressor design requirements.
11. The determining method according to claim 1, further comprising: By inputting the design parameters of the compressor into the trained neural network model, aerodynamic calculations are performed to obtain the performance parameters of the compressor.
12. A model determining device, comprising: The generation module is configured to generate datasets based on compressor-based experimental data. The extraction module is configured to extract constraint rules based on the compressor's design rules; The filtering module is configured to input the dataset and the constraint rules into a neural network model to filter out data in the dataset that satisfy the constraint rules through the neural network model, wherein the neural network model is used to perform aerodynamic calculations for the compressor; The training module is configured to train the neural network model based on the data that satisfies the constraint rules, so as to obtain a trained neural network model.
13. A model determining device, comprising: processor; as well as A memory coupled to the processor is used to store instructions that, when executed by the processor, cause the processor to perform the method for determining the model as described in any one of claims 1 to 11.
14. A computer-readable storage medium having a computer program stored thereon, wherein, When the program is executed by the processor, it implements the method for determining the model as described in any one of claims 1 to 11.
15. A computer program product comprising instructions that, when executed by a processor, cause the processor to perform the method for determining a model according to any one of claims 1 to 11.