CN-121978921-A - Prediction control method and device for aero-engine with online reconstruction of prediction model

CN121978921ACN 121978921 ACN121978921 ACN 121978921ACN-121978921-A

Abstract

The invention discloses an aeroengine prediction control method for online reconstruction of a prediction model. The method comprises the steps of carrying out online reconstruction on a prediction output model, monitoring the height, mach number and control quantity and the model deviation degree of the prediction output model, updating a limited data set when the height, mach number and control quantity change, determining the number N i,j of the minimum data set required by subspace identification in the current state according to a nonlinear function relation established in advance in an offline mode, extracting the first N i,j data at the current moment from the limited data set, and if the model deviation degree exceeds a preset threshold value, calculating a Hankel matrix again by using the N i,j data, and utilizing the Hankel matrix to reconstruct the prediction output model online. The invention also discloses an aeroengine prediction control device for online reconstruction of the prediction model. The method can remarkably improve the accuracy, robustness and control effectiveness of predictive control in the full working condition range.

Inventors

CHEN QIAN
HU JUNHAO
YANG TAO
Shi Haolan
SHENG HANLIN

Assignees

南京航空航天大学

Dates

Publication Date: 20260505
Application Date: 20260114

Claims (8)

1. The prediction control method for the on-line reconstruction of the prediction model comprises the steps of carrying out prediction control on the aircraft engine based on subspace identification on the basis of a prediction output model, carrying out on-line reconstruction on the prediction output model by using the method, wherein the method comprises the steps of monitoring the height, mach number and control quantity of the aircraft engine in real time, simultaneously monitoring the model deviation degree of the prediction output model in real time, wherein the model deviation degree is the average value of errors between the prediction output quantity and the actual output quantity of the prediction output model in a backtracking time domain, when the height, mach number and control quantity of the aircraft engine change, updating a limited dataset at the current moment by using the height, mach number and control quantity at the current moment, determining the minimum dataset number N i,j required by subspace identification corresponding to the current state according to the nonlinear function relation between the minimum dataset number, the height, the Mach number and the control quantity which are established in advance, and then extracting front N i,j data at the current moment from the updated limited dataset, and if the model deviation degree exceeds a preset threshold value, calculating the prediction output matrix again by using the extracted front N i,j data at the current moment, and reconstructing the Hankel matrix by using the reconstructed prediction output matrix.
2. The prediction model online reconstruction aeroengine prediction control method of claim 1 is characterized in that the nonlinear function relation is built offline in advance by using a method that based on an aeroengine component level model, M different packet point areas are divided according to nonlinear strong and weak characteristics, then under each divided packet point area, the characteristics of strong nonlinearity of an engine under a low thrust level and weak nonlinearity under a high thrust level are considered, sparse division is conducted on each packet point area from a slow vehicle state to a maximum state to obtain P approximate linear state areas, subspace identification is conducted on a limited data set of each state of each divided packet point area in a traversing mode, so that the aim of accurately matching the engine state is achieved, and the minimum data set number N i,j , i=1, 2, M, j=1, 2 required for subspace identification under different packet points and different states is obtained.
3. The method for controlling the prediction of the aeroengine by the on-line reconstruction of the prediction model according to claim 1, wherein if the degree of deviation of the model does not exceed a preset threshold value, the lower triangular matrix R of the current prediction output model is updated by using new input-output data obtained at the current moment.
4. The prediction control method for the aeroengine based on online reconstruction of the prediction model according to claim 1, wherein the control target of the prediction control based on subspace identification is optimal output tracking and minimum oil consumption.
5. The prediction control device of the aeroengine based on subspace identification for the online reconstruction of the prediction model is characterized by comprising an online reconstruction module of the prediction model, wherein the online reconstruction module of the prediction model is used for carrying out real-time monitoring on the height, mach number and control quantity of the aeroengine and simultaneously monitoring the model deviation degree of the prediction output model in real time, the model deviation degree is an average value of errors between the prediction output quantity and the actual output quantity of the prediction output model in a backtracking time domain, when the height, mach number and the control quantity of the aeroengine change, the limited data set at the current moment is updated by the height, mach number and the control quantity at the current moment, the number N i,j of the minimum data set required by subspace identification corresponding to the current state is determined according to the nonlinear function relation between the number of the minimum data set, the height, the Mach number and the control quantity which are established in advance, then the front N i,j data at the current moment are extracted from the updated limited data set, and if the model deviation degree exceeds a preset threshold value, the prediction matrix is calculated again by using the extracted front N i,j , and the Hankel is output to be calculated again, and the prediction matrix is calculated.
6. The prediction model online reconstruction aeroengine prediction control device according to claim 5 is characterized in that the nonlinear function relation is built offline in advance by using a method that based on an aeroengine component level model, M different packet point areas are divided according to nonlinear strong and weak characteristics, then under each divided packet point area, the characteristics of strong nonlinearity of an engine under a low thrust level and weak nonlinearity under a high thrust level are considered, sparse division is conducted on each packet point area from a slow vehicle state to a maximum state to obtain P approximate linear state areas, subspace identification is conducted on a limited data set of each state of each divided packet point area in a traversing mode, so that the aim of accurately matching the engine state is achieved, and the minimum data set number N i,j , i=1, 2, M, j=1, 2 required for subspace identification under different packet points and different states is obtained.
7. The prediction control device for an aeroengine with online reconstruction of a prediction model according to claim 5, further comprising a model updating module for updating the lower triangular matrix R of the current prediction output model by using new input-output data obtained at the current moment when the degree of deviation of the model does not exceed a preset threshold.
8. The prediction model online reconstruction aeroengine prediction control device according to claim 5, wherein the control target of the prediction control based on subspace identification is optimal output tracking and minimum oil consumption.

Description

Prediction control method and device for aero-engine with online reconstruction of prediction model Technical Field The invention relates to an aero-engine predictive control method, and belongs to the technical field of aero-engine control. Background In recent years, predictive control has become the most effective and practical advanced control method for the process, and has become the first choice for model-based advanced performance control system design of aeroengines in various research directions in the aerospace field. However, the key to achieving highly complex, strongly nonlinear predictive control of aircraft engine systems is the acquisition of predictive models. Chen et al (Chen. Research on model-based intelligent control method of aero-engine with high stability and high safety [ D ]. Nanjing university of aviation aerospace, 2023.) propose a direct performance self-adaptive prediction control method of aero-engine, construct a direct performance prediction controller (SIIMPC) based on subspace identification, calculate the past/future input/output Hankel matrix by the principle of subspace identification, deduce the future f-step incremental prediction output model at k moment, and invent two tracking weight on-line self-adaptive adjustment methods to solve the defects brought by adopting constant design parameters of the controller, so as to enhance the flexibility and adaptability of the control system. The SIIMPC controller in the above technical solution adopts a fixed finite data set size, that is, the order of the prediction output model is kept unchanged, and this design solution has a certain technical challenge. In particular, since the nonlinear characteristics of the aero-engine under different working conditions are significantly different, the use of a prediction output model of a fixed order may cause a problem of model mismatch at certain strong nonlinear working points, thereby affecting the control effect. On the other hand, in a state with better linearity, the fixed high-order prediction model may cause unnecessary calculation overhead, resulting in resource waste. Therefore, in order to improve the prediction control performance of the aero-engine in the full envelope and full state, it is necessary to develop a study on the online reconstruction technology of the prediction model. This means that a mechanism needs to be developed, so that the order of the prediction model can be adjusted in real time according to the actual running state of the engine, thereby ensuring that the prediction model can accurately match the state of the actual engine under the working condition of strong nonlinearity or good linearity, and realizing an efficient and stable control effect. Therefore, the control precision can be improved, the utilization efficiency of computing resources can be optimized, and powerful technical support is provided for intelligent control of the aero-engine. Disclosure of Invention The invention aims to overcome the defect that the order of a prediction output model of the existing aeroengine prediction control method is unchanged, and provides an aeroengine prediction control method for online reconstruction of a prediction model, which can adjust the order of the prediction output model in real time according to the dynamic characteristics of an engine under different flight envelope and running states, so that the dynamic matching of a model structure and the real system characteristics is realized, and the accuracy, the robustness and the control effectiveness of the prediction control in a full working condition range are obviously improved. The technical scheme adopted by the invention specifically solves the technical problems as follows: The prediction control method of the aeroengine based on the online reconstruction of the prediction model comprises the steps of carrying out prediction control on the aeroengine based on subspace identification on the basis of a prediction output model, carrying out real-time monitoring on the height, mach number and control quantity of the aeroengine, simultaneously monitoring the model deviation degree of the prediction output model in real time, wherein the model deviation degree is the average value of errors between the prediction output quantity and the actual output quantity of the prediction output model in a backtracking time domain, updating a limited data set at the current moment by using the height, mach number and control quantity of the aeroengine when the height, mach number and control quantity of the aeroengine change, determining the minimum data set number N i,j required by subspace identification corresponding to the current state according to the nonlinear function relation between the minimum data set number, the height, the Mach number and the control quantity which are established in advance, and then extracting first N i,j data at the current moment from the updated limited data set, and c