CN-122014424-A - Multi-working-condition parameter optimization method and device for aero-engine system

CN122014424ACN 122014424 ACN122014424 ACN 122014424ACN-122014424-A

Abstract

The present disclosure relates to a multi-working condition parameter optimization method and apparatus for an aero-engine system, the method comprising obtaining a target optimization task to determine a plurality of components and a plurality of working states of the target aero-engine system; the method comprises the steps of constructing an optimization model with a neural network structure based on a plurality of components and a plurality of working states of a target aeroengine system, training an initial optimization model based on a preset first optimization strategy and a preset second optimization strategy in each model iteration process until the trained optimization model meets preset conditions to obtain a trained optimization model, and inputting a target adjustment direction into the optimization model to obtain an optimization result. The method solves the problem of unified optimization of heterogeneous components by adopting an alternative optimization strategy, breaks through the limitation of the traditional single modeling paradigm, coordinates and processes different types of model components under a unified computing framework, and provides a brand-new technical path for multi-working condition optimization of a complex engineering system.

Inventors

FENG XUDONG
XU QIANNAN
WU FENG
ZHANG HAI
HAN YU

Assignees

中国航发四川燃气涡轮研究院

Dates

Publication Date: 20260512
Application Date: 20260116

Claims (10)

1. A multi-operating parameter optimization method for an aircraft engine system, the method comprising: acquiring a target optimization task aiming at a target aero-engine system, and determining a plurality of parts and a plurality of working states of the target aero-engine system; An optimization model with a neural network structure is constructed based on a plurality of components and a plurality of working states of the target aero-engine system, the optimization model comprises a working condition neural network corresponding to each working state, each working condition neural network comprises a plurality of component layers, and each component layer corresponds to different components; training the initial optimization model based on a first preset optimization strategy and a second preset optimization strategy in each model iteration process until the trained optimization model meets preset conditions, so as to obtain a trained optimization model; and determining a target adjustment direction aiming at the target aeroengine system according to the target optimization task, and inputting the target adjustment direction into the optimization model to obtain an optimization result, wherein the optimization result comprises parameter adjustment amounts of components required for meeting the target adjustment direction.
2. The method according to claim 1, wherein the method further comprises: Dividing a plurality of components of the target aeroengine system into a micro-component and a non-micro-component according to a preset classification strategy, wherein the first optimization strategy corresponds to the micro-component, and the second optimization strategy corresponds to the non-micro-component.
3. The method of claim 2, wherein the classification policy comprises at least one of: Dividing a part with continuously-changed state parameters along with input into micro parts, and dividing a part with abrupt change of the state parameters into non-micro parts; dividing the parts of which the part behaviors are influenced by the continuous physical field into micro parts, and dividing the parts of which the part behaviors are influenced by discrete events into non-micro parts; dividing a part with geometric features exhibiting continuous curvature change into micro-parts, and dividing a part with geometric structures with discontinuous boundaries into non-micro-parts; the components whose output responses are in a linear or smooth nonlinear relationship are divided into micro-components, and the components whose output responses have a threshold effect are divided into non-micro-components.
4. The method of claim 2, wherein training an initial optimization model according to the first optimization strategy comprises: Network parameters of component layers of all non-micro components are fixed, and gradient optimization is performed on the micro components by utilizing a multi-working state back propagation algorithm.
5. The method of claim 2, wherein training an initial optimization model according to the second optimization strategy comprises: Fixing network parameters of component layers of all the micro-components, and performing gradient-free optimization on the non-micro-components by adopting a Bayesian optimization algorithm based on a Gaussian process.
6. The method according to claim 1, wherein the method further comprises: in each model iteration process, calculating a loss function value based on the dynamic balance coefficient; And under the condition that the loss function value is smaller than a preset threshold value, obtaining a trained optimization model.
7. A multiple operating parameter optimizing device for an aircraft engine system, the device comprising: The task acquisition module is used for acquiring a target optimization task aiming at a target aero-engine system and determining a plurality of parts and a plurality of working states of the target aero-engine system; The model construction module is used for constructing an optimization model with a neural network structure based on a plurality of components and a plurality of working states of the target aero-engine system, the optimization model comprises a working condition neural network corresponding to each working state, each working condition neural network comprises a plurality of component layers, and each component layer corresponds to different components; the model training module is used for training the initial optimization model based on a first preset optimization strategy and a second preset optimization strategy in each model iteration process until the trained optimization model meets preset conditions, so as to obtain a trained optimization model; and the parameter optimization module is used for determining a target adjustment direction for the target aeroengine system according to the target optimization task, inputting the target adjustment direction into the optimization model, and obtaining an optimization result, wherein the optimization result comprises parameter adjustment amounts of components required for meeting the target adjustment direction.
8. A multiple operating parameter optimizing device for an aeroengine system, comprising a memory, a processor and a computer program stored on the memory, characterized in that the processor executes the computer program to carry out the steps of the method according to any one of claims 1 to 6.
9. A non-transitory computer readable storage medium having stored thereon a computer program, characterized in that the computer program when executed by a processor realizes the steps of the method according to any of claims 1 to 6.
10. A computer program product comprising a computer program, or a non-transitory computer readable storage medium carrying a computer program, characterized in that the computer program when executed by a processor implements the steps of the method according to any one of claims 1 to 6.

Description

Multi-working-condition parameter optimization method and device for aero-engine system Technical Field The disclosure relates to the technical field of engine optimization, in particular to a multi-working-condition parameter optimization method and device for an aeroengine system. Background In an aeroengine system, how to process the complex input-output relation of a component layer and the coupling optimization problem of prediction and a system layer through an artificial intelligence (ARTIFICIAL INTELLIGENCE, AI) architecture specifically comprises the steps of utilizing a nonlinear input-output mapping relation which cannot be accurately described by a traditional formula in a part of key components of A learning, simultaneously establishing a parameter optimization algorithm capable of processing the complex coupling influence of a front component on a rear component, realizing global parameter optimization under various working conditions, and ensuring that a set of optimal parameter combination capable of adapting to various working conditions is found, thereby solving the technical bottleneck faced by the traditional engine modeling method in processing the nonlinear coupling relation among the components and optimizing the parameters of multiple working conditions. Disclosure of Invention In view of this, the present disclosure proposes a multi-operating parameter optimization method and apparatus for an aero-engine system. According to one aspect of the disclosure, a multi-working-condition parameter optimization method for an aeroengine system is provided, the method comprises the steps of obtaining a target optimization task for the target aeroengine system, determining a plurality of components and a plurality of working states of the target aeroengine system, constructing an optimization model with a neural network structure based on the components and the working states of the target aeroengine system, wherein the optimization model comprises a working-condition neural network corresponding to each working state, each working-condition neural network comprises a plurality of component layers, each component layer corresponds to a different component, training an initial optimization model until the trained optimization model meets preset conditions based on a preset first optimization strategy and a preset second optimization strategy in each model iteration process, obtaining a trained optimization model, determining a target adjustment direction for the target aeroengine system according to the target optimization task, and inputting the target adjustment direction into the optimization model to obtain an optimization result, wherein the optimization result comprises parameter adjustment amounts of components required for meeting the target adjustment direction. Therefore, the engine system is reconfigured into the neural network form, the unified optimization problem of heterogeneous components is solved by adopting an alternative optimization strategy, the limitation of the traditional single modeling paradigm is broken through, different types of model components are cooperatively processed under a unified computing frame, and a brand new technical path is provided for multi-condition optimization of the complex engineering system. According to the multi-working-condition parameter optimization method, through a layering modeling and mixing optimization strategy of micro/non-micro components, the difficult problem of parameter coupling and nonlinear optimization under the multi-working conditions of an aeroengine is solved, and the component layers of the working-condition neural network are designed in a differentiated mode according to the characteristic difference of the micro/non-micro components in the engine system. For example, the micro-components (such as a compressor and a turbine) adopt a traditional neural network structure, and a continuous mapping relation is constructed by linear transformation and a micro-activatable function (such as a ReLU), so that the whole micro-components can be kept. For example, non-micro components (such as an ignition system and a deflation valve) can be embedded into a mixed structure of a decision tree and a neural network (such as a micro-adaptive neural tree ANT), discrete threshold logic is converted into continuous probability output through a soft decision path, and the problem of insufficient modeling capability of the traditional neural network on discrete events is solved. The layering processing ensures that the optimization model can accurately capture the step characteristics of the non-microminiaturizable component while maintaining the global microminiaturization, and compared with a single neural network model, the parameter prediction error under multiple working conditions is reduced. And a closed-loop training mechanism is formed by a first optimization strategy (data driving) and a second optimization strategy (physical c