A model predictive control method for aircraft engines based on spatiotemporal learning

Through the model prediction and control method based on space-time learning, the space-time Gaussian process and Kalman filtering technology are used to solve the problems of spatiotemporal changes and real-time in the prediction and control of aero engine models, the control performance and constraint satisfaction capabilities are improved, and dynamic information prediction and control are realized Real-time optimization of input.

CN115202207BActive Publication Date: 2024-11-05NORTHEASTERN UNIV CHINA
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210833293.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-07-14
Publication Date
2024-11-05
Estimated Expiration
2042-07-14

AI Technical Summary

Technical Problem

The prior art is difficult to effectively solve the problem of model uncertainty and real-time in the prediction control of space-time changes in aero engine models, especially when the system is complex and external disturbances exist, it is difficult for traditional methods to achieve high-performance real-time optimization control.

Method used

Using a model prediction and control method based on space-time learning, the state space expression is constructed by establishing nonlinear dynamic equations and real-time operating states of aero engines, using technologies such as spatiotemporal Gaussian processes and Kalman filtering, and optimizing control inputs to improve real-time and control. performance.

Benefits of technology

It improves the performance and real-time performance of predictive control of aircraft engine models, enhances the processing ability of spatial and temporal changes in model uncertainty, satisfies the balance between optimal control performance and system constraints, and reduces the computational complexity.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115202207B_ABST
    Figure CN115202207B_ABST
Patent Text Reader

Abstract

The invention discloses an aircraft engine model predictive control method based on spatiotemporal learning, which belongs to the field of aircraft engine high-performance control and data-driven system control. The method comprises the following steps: establishing a nonlinear dynamic equation of the aircraft engine and the real-time operation state of the aircraft engine; adopting a model predictive control algorithm based on spatiotemporal learning to realize the prediction of dynamic information in the actual operation process of the aircraft engine and improve the real-time performance of solving the control input. The invention improves the performance of the aircraft engine model predictive control. The invention designs a spatiotemporal learning method based on Gaussian process under the framework of model predictive control, that is, learning the unknown disturbance in the spatiotemporal field, increasing the accuracy of the prediction model, improving the control performance and constraint satisfaction capability, and performing state space conversion on the Gaussian process to reduce the model calculation complexity so as to meet the real-time performance of the designed controller.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of high-performance control of aircraft engines and data-driven system control, and specifically relates to an aircraft engine model predictive control method based on spatiotemporal learning, which aims to solve the problem of high-performance real-time operation optimization control of aircraft engines in the presence of model uncertainty with spatiotemporal variations and input constraints. Background Art

[0002] The rapid development of modern aircraft, especially military aircraft for high-tech warfare, has put forward a series of higher requirements for future high-performance aircraft engines: high thrust-to-weight ratio, low fuel consumption, high reliability, long life and light weight. Fully tapping the performance potential of high-performance aircraft engines and dynamically optimizing the performance of aircraft engines are of great significance to improving the level of aviation science and technology, serving national defense and even the entire national economy.

[0003] Model predictive control is an efficient dynamic optimization strategy. Model predictive control of aircraft engines is a complex problem in the field of aircraft engine optimization. On the one hand, the particularity and complexity of the working process of the aircraft engine system make it difficult to accurately predict the system dynamic information during the actual operation of the aircraft engine; on the other hand, due to the system characteristics of the aircraft engine, there are high requirements for the real-time solution of the model predictive control algorithm.

[0004] There are two main reasons why the dynamic information of aircraft engine systems is difficult to predict: first, there is model uncertainty caused by external disturbances, measurement noise, etc.; second, the hyperparameters that determine the statistical characteristics of model uncertainty may change with time and space. To address the problem of system dynamic uncertainty, traditional methods mainly include: (1) For known or partially known system dynamics, express it as a nominal system with disturbances, use prior knowledge to model the disturbance as random noise with boundary constraints or known probability density, and use robust model predictive control, Tube-model predictive control or stochastic model predictive control to process the random noise. However, such methods require the statistical characteristics of unknown noise to be clear, which is difficult for aircraft engine systems with complex operating environments; (2) Use neural networks, Gaussian process regression and other methods to learn model uncertainty. However, most of these models do not take into account the problem of dynamic changes in the statistical characteristics of model uncertainty, and are not suitable for online solving the problems caused by aircraft engine system uncertainty; (3) Use spatiotemporal learning methods to train dynamically changing model uncertainty. However, such existing methods have not improved the aircraft engine model predictive control method and are not suitable for aircraft engine systems with complex system models and high real-time requirements. Therefore, new technical methods are needed to design appropriate model predictive control algorithms to address the problem of spatiotemporal variations in the statistical characteristics of model uncertainty.

[0005] Traditional solutions to the real-time problem of model predictive control methods mainly include: (1) directly using existing offline optimization toolkits to solve; (2) for nonlinear systems, linearizing the system near the operating point and solving it through linear quadratic regulator (LQR) or linear matrix inequality (LMI); (3) solving it through adaptive dynamic programming (ADP); (4) solving it using heuristic algorithms such as genetic algorithm, ant colony algorithm, particle swarm algorithm, etc. However, the solution method of model predictive control problem based on spatiotemporal learning and the trade-off relationship between optimal control performance and system constraint satisfaction are still worthy of further study. Summary of the invention

[0006] In order to solve the above problems, the present invention provides a technical solution adopted by the present invention: an aircraft engine model predictive control method based on spatiotemporal learning, comprising the following steps:

[0007] S10: Establishing the nonlinear dynamic equation of the aircraft engine and the real-time operating status of the aircraft engine;

[0008] S20: Based on the nonlinear dynamic equations of the aircraft engine and the real-time operating status of the aircraft engine, a model predictive control algorithm based on spatiotemporal learning is adopted to realize the prediction of dynamic information during the actual operation of the aircraft engine and improve the real-time performance of the control input solution.

[0009] Further: Based on the nonlinear dynamic equation of the aircraft engine and the real-time operation state of the aircraft engine, the model predictive control algorithm based on spatiotemporal learning is adopted to realize the prediction of dynamic information in the actual operation process of the aircraft engine and improve the real-time performance of the control input solution as follows:

[0010] S201: Establishing an aircraft engine small deviation gain model with the current actual input of the aircraft engine as the steady-state point, after obtaining the linear expressions of the aircraft engine state variables and output variables, using the state space expression to construct the control model of the aircraft engine, and at the same time, given the initial state x0 of the system, constructing the state space expression of the space-time Gaussian process, offline calculating the observation matrix M and the feedback matrix K, and selecting the quadratic weights Q, R;

[0011] S202: Get aircraft engine t k The actual state measurement value y at the moment k As well as the expected state variables, calculate the estimation error, and measure the external environment parameters p k ;

[0012] S203: Calculate t k The deviation between the measured value of the aircraft engine system state and the nominal model at the moment, and the deviation data is compared with the data set I kMake updates;

[0013] S204: Use the offline calculated measurement matrix M to perform Kalman filtering on the system and calculate the augmented system state and the error covariance Σ k Perform calculations;

[0014] S205: Based on the estimated system state and error covariance, the infinite time domain optimal control problem is approximated to solve the finite time domain optimal control {u 0|k ,…,u N-1|k}, the output of the aircraft engine controller is used as the control input u of the real system k =u 0|k ;

[0015] S206: Update the iteration sequence k=k+1, and return to S202 until the aircraft engine stops running.

[0016] Further: The expression of the control model of the aircraft engine is as follows:

[0017]

[0018] In the formula, A, B, C are the discretized parameter matrices, k is the discrete time step, They are respectively represented as the system state, control input, measurement output and operating condition at the time instant, and the unknown dynamic g p,k It is expressed as the influence of external disturbance on the system at time k and hyperparameter p, w k is the process noise, v k To measure noise.

[0019] Further: the process of solving the expression of the control model of the aircraft engine is as follows:

[0020] Assume a vector variable and a continuous mapping function A Gaussian process is described as:

[0021]

[0022] Where, μ(ξ), They are respectively expressed as the mean and covariance of the corresponding function value f(ξ). In order to reduce the problem of excessive computational requirements caused by long-term control, we first Perform Fourier transform to obtain the corresponding n-order Fourier domain stochastic differential equation:

[0023]

[0024] In the formula, is a constant coefficient, w(iω p ,t) The driving function is Gaussian white noise;

[0025] By solving the inverse Fourier transform in the above equation, it can be converted from the spatial Fourier domain to the spatial domain. Using control theory, the corresponding state space expression can be obtained:

[0026]

[0027] The state variables are The system matrix is ​​expressed as:

[0028]

[0029] Therefore, given the state-space equation, the power spectral density of g(p,t) can be calculated as:

[0030]

[0031] where σ w is the power spectral density of Gaussian white noise, and H(-iω) is the system transfer function. In addition, we can obtain a stable kernel function based on the inverse Fourier transform:

[0032]

[0033] By combining the mechanical model of the aircraft engine, we can obtain an augmented state space model:

[0034]

[0035] Where, the system parameter matrix is:

[0036]

[0037] Further: the process of improving the real-time performance of the control input solution is as follows:

[0038] The controller parameterization is used to calculate the control input sequence. The asymptotically stable feedback matrix K and the mean square stable measurement matrix M are calculated offline to achieve the following controller parameterization:

[0039]

[0040] in Expressed as χ k A priori estimate of χ k A posteriori estimation, based on the definition of controller parameterization, we can obtain the corresponding i-th step prediction state With control input u i|k :

[0041]

[0042] Therefore, the predicted state sequence and the estimated error series can be expressed as:

[0043]

[0044] in, Consider the estimation error σ k The covariance matrix of The update formula of the covariance matrix is:

[0045]

[0046] According to the above analysis, the corresponding predictive control input sequence u can be obtained k , the estimated state sequence and the estimated error series δ k :

[0047]

[0048] in as well as

[0049]

[0050] Further: the aero-engine objective function is:

[0051] The infinite time domain is reconstructed as:

[0052]

[0053] The terminal term can be defined as:

[0054]

[0055] Terminal term, make the following definition:

[0056]

[0057] Through the above definition, we get:

[0058]

[0059] Among them, the parameter matrix in the formula can be described as According to P k The definition of , infers:

[0060]

[0061] Therefore, the terminal term It can be calculated by tr(ZP k ) is obtained, where At the same time k represents the solution of the stochastic Lyapunov equation.

[0062]

[0063] An aircraft engine model predictive control system based on spatiotemporal learning, comprising:

[0064] Initial setting module: used to establish an aircraft engine small deviation gain model with the current actual input of the aircraft engine as the steady-state point. After obtaining the linear expressions of the aircraft engine state variables and output variables, the state space expression is used to build the control model of the aircraft engine. At the same time, given the initial state x0 of the system, the state space expression of the space-time Gaussian process is constructed, the observation matrix M and the feedback matrix K are calculated offline, and the appropriate quadratic weights Q, R are selected;

[0065] State measurement module: used to obtain the aircraft engine t k The actual state measurement value y at the moment k As well as the expected state variables, calculate the estimation error, and measure the external environment parameters p k ;

[0066] Cumulative learning module: used to configure the calculation of t k The deviation between the measured value of the system state and the nominal model at the moment, and the deviation data is compared with the data set I k Make updates;

[0067] Real-time observation module: used to configure the system to use the offline calculated observation matrix M to perform Kalman filtering on the system and to calculate the augmented system state. and the error covariance Σ k Perform calculations;

[0068] Model Predictive Control Module: It is used to configure the system state and error covariance based on the estimated system state, approximate the infinite time domain optimal control problem, and solve the finite time domain optimal control {u 0|k ,…,u N-1|k}, the controller output is used as the control input u of the real system k =u 0|k ;

[0069] The outer loop module is configured to set k=k+1 and execute step S20-step S60 cyclically until the aircraft engine reaches an initially set system state.

[0070] The present invention provides an aircraft engine model predictive control method based on spatiotemporal learning. By changing the structure of standard spatiotemporal Gaussian process learning and using the state space representation of the spatiotemporal Gaussian process, the problem of expensive calculation cost in the long-term control time domain is greatly reduced. At the same time, in terms of operation optimization control, by using Kalman filtering or Bayesian filtering, the state and error covariance of the aircraft engine system are estimated, and the optimal control problem of the spatiotemporal variation of model uncertainty is solved, thereby improving the control performance and constraint satisfaction of the aircraft engine.

[0071] This method has the following advantages: the present invention improves the performance of aircraft engine model predictive control. The present invention designs a space-time learning method based on Gaussian process under the framework of model predictive control, that is, learning unknown disturbances in the space-time field, increasing the accuracy of the prediction model, improving control performance and constraint satisfaction capabilities, and at the same time performing state space transformation on the Gaussian process to reduce the complexity of model calculation to meet the real-time performance of the designed controller.

[0072] At the same time, by constructing terminal constraints on finite time domain optimal control, through Lyapunov stability theorem and combined with mathematical induction, the parameter conditions that meet the iterative feasibility and closed-loop stability of the system are obtained, and the control performance is analyzed.

[0073] Therefore, designing a new aircraft engine model predictive control method is of great significance for improving the dynamic information prediction capability of aircraft engines, enhancing the real-time performance of the algorithm, and achieving a balance between optimal control performance and system state constraint satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0075] Figure 1 It is a flow chart of an aircraft engine model predictive control method based on spatiotemporal learning according to an embodiment of the present invention;

[0076] Figure 2 It is a schematic diagram of a framework of an aircraft engine model predictive control method based on spatiotemporal learning according to an embodiment of the present invention;

[0077] Figure 3 It is a control block diagram of an aircraft engine model predictive control method based on spatiotemporal learning according to an embodiment of the present invention;

[0078] Figure 4(a) is a schematic diagram of an aircraft engine compressor control system according to an embodiment of the present invention; (b) is an overview diagram of an aircraft engine simulation closed-loop system according to an embodiment of the present invention;

[0079] Figure 5 It is a schematic diagram of measurement positions of external disturbances within the flight envelope of an embodiment of the present invention;

[0080] Figure 6 (a) is an estimated comparison curve of external disturbance at a measurable node of an aircraft engine according to an embodiment of the present invention, and (b) is an estimated comparison curve of external disturbance at an unmeasurable node of an aircraft engine according to an embodiment of the present invention;

[0081] Figure 7 (a) is a comparison curve of mass flow of an aircraft engine under external disturbance according to an embodiment of the present invention; (b) is a comparison curve of throat opening of an aircraft engine under external disturbance according to an embodiment of the present invention;

[0082] Figure 8 It is a multiple comparison curve of aircraft engine compressor variables under different working conditions of an embodiment of the present invention. DETAILED DESCRIPTION

[0083] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0084] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. The following description of at least one exemplary embodiment is actually only illustrative and is by no means intended to limit the present invention and its application or use. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0085] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0086] Unless otherwise specifically stated, the relative arrangement of the parts and steps described in these embodiments, the numerical expressions and numerical values ​​do not limit the scope of the present invention. At the same time, it should be clear that, for ease of description, the sizes of the various parts shown in the drawings are not drawn according to the actual proportional relationship. The technology, methods and equipment known to ordinary technicians in the relevant field may not be discussed in detail, but in appropriate cases, the technology, methods and equipment should be regarded as part of the authorization specification. In all examples shown and discussed here, any specific value should be interpreted as merely exemplary, rather than as a limitation. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar numbers and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further discussed in subsequent drawings.

[0087] In the description of the present invention, it is necessary to understand that the directions or positional relationships indicated by directional words such as "front, back, up, down, left, right", "lateral, vertical, perpendicular, horizontal" and "top, bottom" are usually based on the directions or positional relationships shown in the drawings. They are only for the convenience of describing the present invention and simplifying the description. Unless otherwise specified, these directional words do not indicate or imply that the device or element referred to must have a specific direction or be constructed and operated in a specific direction. Therefore, they cannot be understood as limiting the scope of protection of the present invention: the directional words "inside and outside" refer to the inside and outside relative to the contours of each component itself.

[0088] For ease of description, spatially relative terms such as "above", "above", "on the upper surface of", "above", etc. may be used here to describe the spatial positional relationship between a device or feature and other devices or features as shown in the figure. It should be understood that spatially relative terms are intended to include different orientations of the device in use or operation in addition to the orientation described in the figure. For example, if the device in the accompanying drawings is inverted, the device described as "above other devices or structures" or "above other devices or structures" will be positioned as "below other devices or structures" or "below their position devices or structures". Thus, the exemplary term "above" can include both "above" and "below". The device can also be positioned in other different ways (rotated 90 degrees or in other orientations), and the spatially relative descriptions used here are interpreted accordingly.

[0089] In addition, it should be noted that the use of terms such as "first" and "second" to limit components is only for the convenience of distinguishing the corresponding components. If not otherwise stated, the above terms have no special meaning and therefore cannot be understood as limiting the scope of protection of the present invention.

[0090] A method for predictive control of an aircraft engine model based on spatiotemporal learning, characterized in that it comprises the following steps:

[0091] S10: Establishing the nonlinear dynamic equation of the aircraft engine and the real-time operating status of the aircraft engine;

[0092] S20: Based on the nonlinear dynamic equations of the aircraft engine and the real-time operating status of the aircraft engine, a model predictive control algorithm based on spatiotemporal learning is adopted to realize the prediction of dynamic information during the actual operation of the aircraft engine and improve the real-time performance of the control input solution.

[0093] Furthermore, the process of using a model predictive control algorithm based on spatiotemporal learning based on the nonlinear dynamic equation of the aircraft engine and the real-time operation state of the aircraft engine to realize the prediction of dynamic information during the actual operation of the aircraft engine and improve the real-time performance of the control input solution is as follows:

[0094] S201: Establishing an aircraft engine small deviation gain model with the current actual input of the aircraft engine as the steady-state point, after obtaining the linear expressions of the aircraft engine state variables and output variables, using the state space expression to construct the aircraft engine control model, and at the same time, given the initial state x0 of the aircraft engine system, constructing the state space expression of the space-time Gaussian process, offline calculating the observation matrix M and the feedback matrix K, and selecting appropriate quadratic weights Q, R;

[0095] S202: Get aircraft engine t k The actual state measurement value y at the moment k As well as the expected state variables, calculate the estimation error, and measure the external environment parameters p k ;

[0096] S203: Calculate t k The deviation between the measured value of the aircraft engine system state and the nominal model at the moment, and the deviation data is compared with the data set I k Make updates;

[0097] S204: Use the offline calculated observation matrix M to perform Kalman filtering on the system to obtain the augmented aircraft engine system state and the error covariance Σ k Perform calculations;

[0098] S205: Based on the estimated aircraft engine system state and error covariance, approximate the infinite time domain optimal control problem and solve the finite time domain optimal control {u 0|k ,…,u N-1|k}, the output of the aircraft engine controller is used as the control input u of the real system k =u 0|k;

[0099] S206: Set the iteration sequence to update k=k+1, and return to S202 until the aircraft engine reaches the initially set system state.

[0100] In order to more clearly illustrate the aircraft engine model predictive control method based on spatiotemporal learning of the present invention, the steps in an embodiment of the method of the present invention are described in detail below with reference to the accompanying drawings.

[0101] 1. Construction of prediction model and spatiotemporal learning

[0102] Step A100, establish an aircraft engine small deviation gain model with the current actual input of the aircraft engine as the steady-state point, and after obtaining the linear expressions of the aircraft engine state variables and output variables, use the state space expression to construct the control model of the aircraft engine. The expression of the control model of the aircraft engine is as follows:

[0103]

[0104] In the formula, A, B, C are the discretized parameter matrices, k is the discrete time step, They represent the system state, control input, measurement output and operating condition at the time. At the same time, we assume that the process noise w k With the measurement noise v k is independent and identically distributed Gaussian noise:

[0105]

[0106] Unknown dynamics p,k It is expressed as the influence of external disturbance on the system at time k and hyperparameter p. The controller can obtain the state mean at the initial time Initial state covariance:

[0107]

[0108] According to the state space expression of the system, we give the common infinite time domain quadratic expression as the optimal control objective function:

[0109]

[0110] In the formula, β∈(0,1) is the discount factor, Q>0, R>0 is the semi-positive definite weight matrix. In order to ensure the safe and stable operation of the aircraft engine, we consider the following chance constraint:

[0111]

[0112] That is, using the state xk The conditional density and known noise distribution are used to strengthen the state constraints. Similarly, we also give the constraints corresponding to the control input:

[0113]

[0114] Step A200, construct an unknown dynamic model in the space-time field based on the indirect measurement data set. Space-time learning usually uses a limited set of indirect measurement space-time data to perform statistical inference. In the present invention, we consider statistical inference of model uncertainty in the space-time field based on the statistical characteristics of the dynamic environment and a limited set of indirect measurements.

[0115] The process of solving the expression of the control model of the aircraft engine is as follows:

[0116] Assume a vector variable and a continuous mapping function The Gaussian process can be described as

[0117]

[0118] Where, μ(ξ), are respectively expressed as the mean and covariance of the corresponding function value f(ξ). In order to reduce the problem of excessive computational requirements caused by long-term control, we first Perform Fourier transform to obtain the corresponding n-order Fourier domain stochastic differential equation:

[0119]

[0120] In the formula, a0(iω p ),...,a nz-1 (iω p ) is a constant coefficient, w(iω p ,t) The driving function is Gaussian white noise.

[0121] By solving the inverse Fourier transform in the above equation, it can be converted from the spatial Fourier domain to the spatial domain. Using control theory, the corresponding state space expression can be obtained:

[0122]

[0123] The state variables are The system matrix can be expressed as:

[0124]

[0125] Therefore, given the state-space equation, the power spectral density of g(p,t) can be calculated as:

[0126]

[0127] where σ w is the power spectral density of Gaussian white noise, and H(-iω) is the system transfer function. In addition, we can obtain a stable kernel function based on the inverse Fourier transform:

[0128]

[0129] By combining the mechanical model of the aircraft engine, we can obtain an augmented state space model:

[0130]

[0131] In the formula, the system parameter matrix is:

[0132]

[0133] Step A300, in order to clearly describe the probability distribution of the predicted state and the control input, the controller parameterization is used to calculate the control input sequence, and the controller parameterization is implemented by offline calculating the asymptotically stable feedback matrix K and the mean square stable measurement matrix M to achieve the following controller parameterization:

[0134]

[0135] in: Expressed as χ k A priori estimate of χ k A posteriori estimation, based on the definition of controller parameterization, we can obtain the corresponding i-th step prediction state With control input u i|k :

[0136]

[0137] Therefore, the predicted state sequence and the estimated error series can be expressed as

[0138]

[0139] in, Consider the estimation error σ k The covariance matrix of The update formula of the covariance matrix is:

[0140]

[0141] According to the above analysis, we can obtain the corresponding predictive control input sequence u k , the estimated state sequence and the estimated error series δ k :

[0142]

[0143] in as well as

[0144]

[0145] Step A400, the process of obtaining the real-time objective function by solving the control input is as follows: Since the infinite time domain objective function is computationally difficult to handle, in this embodiment, a suitable objective function approximation form is considered. The infinite time domain can be reconstructed as:

[0146]

[0147] The terminal term can be defined as:

[0148]

[0149] To better describe the terminal term, we make the following definition:

[0150]

[0151] According to the above definition, we can get

[0152]

[0153] Among them, the parameter matrix in the formula can be described as According to P k From the definition of , we can infer that:

[0154]

[0155] Therefore, the terminal term c N (x i|N ) can be calculated by calculating tr(ZP k ) is obtained, where At the same time k Represents the solution to the stochastic Lyapunov equation:

[0156]

[0157] Step A500, using the controller parameterized state prediction sequence and error covariance, approximate the intractable chance constraints to obtain the corresponding state space linear constraints. The tightened forecast state constraints can be obtained:

[0158]

[0159] According to the tightening constraints, chance constraints can be satisfied such as:

[0160]

[0161] The chance constraint can be explicitly expressed using the quantile function of the standard normal distribution:

[0162]

[0163] At the same time, the tightening constraint on the average of the estimated states can be defined by the predicted reachable set at step i:

[0164]

[0165] According to the approximated objective function and constraints, a convex optimization problem is constructed to approximate the original optimal control problem, and the quadratic programming algorithm is used for rolling horizon optimization. After solving the control input sequence of the optimal control problem, the first element of the sequence is taken as the actual control output in the next control cycle.

[0166] Embodiment, a method for predictive control of an aircraft engine model based on spatiotemporal learning, specifically comprising the following steps:

[0167] Step S10: Given the initial state x0 of the system, construct the state space expression of the space-time Gaussian process, calculate the observation matrix M and the feedback matrix K offline, and select appropriate quadratic weights Q, R;

[0168] In this embodiment, the state space transformation of the spatiotemporal Gaussian process is first performed, such as the above-mentioned step A200.

[0169] Step S20: Obtain aircraft engine t k The actual state measurement value y at the moment k As well as the expected state variables, calculate the estimation error, and measure the external environment parameters p k ;

[0170] In this embodiment, based on the state estimation value at the previous moment, combined with the nominal model of the aircraft engine, the model uncertainty deviation under the current working condition is calculated, and the spatiotemporal learning data set is updated. In the present invention, in order to further improve the stability and accuracy of the control rate, the data set obtained in the general early operation is not directly applied to the model predictive controller, but the hyperparameters in the spatiotemporal Gaussian process are trained to obtain a more suitable kernel function.

[0171] Step S30: Calculate t k The deviation between the measured value of the system state and the nominal model at the moment, and the deviation data is compared with the data set I k Make updates;

[0172] In this embodiment, the aircraft engine state space model is used for updating.

[0173] Step S40: Use the offline calculated measurement matrix M to perform Kalman filtering on the system and calculate the augmented system state and the error covariance Σ k Perform calculations;

[0174] In this embodiment, a parameterized model predictive controller is used to update the system state and error co-square, such as step A300 above.

[0175] Step S50: Based on the estimated system state and error covariance, approximate the infinite time domain optimal control problem to solve the finite time domain optimal control {u 0|k ,…,u N-1|k}, the controller output is used as the control input u of the real system k =u 0|k ;

[0176] In this embodiment, based on the spatiotemporal learning method after the state space display, the model uncertainty at the current moment is estimated, and the optimal control rate at the current moment is solved by the quadratic programming algorithm in combination with the infinite time domain objective function and the opportunity constraint after the processing, and acts on the aircraft engine, such as the above step A500.

[0177] Step S60: let k=k+1, and execute steps S20 to S60 in a loop until the aircraft engine reaches the initially set system state.

[0178] In this embodiment, the actual state and operating conditions of the aircraft engine from the initial state to the system setting value are cyclically acquired, the corresponding optimal control rate is calculated, and acts on the aircraft engine until the set system state is reached.

[0179] A second embodiment of the present invention is an aircraft engine model prediction control system based on spatiotemporal learning, such as Figure 2 As shown, it includes: an initial setting module 100, a state measurement module 200, a cumulative learning module 300, a real-time observation module 400, a model predictive control module 500 and an outer loop module 600;

[0180] The initial setting module 100 is configured to construct a state space expression of a spatiotemporal Gaussian process given an initial state x0 of the system, calculate the observation matrix M and the feedback matrix K offline, and select appropriate quadratic weights Q, R;

[0181] The state measurement module 200 is configured to obtain the aircraft engine t k The actual state measurement value y at the moment k(high-pressure rotor speed, low-pressure rotor speed, etc.) and expected state variables, calculate the estimation error, and measure the external environmental parameters p k (flight altitude, flight Mach number, etc.);

[0182] The cumulative learning module 300 is configured to calculate t k The deviation between the measured value of the system state and the nominal model at the moment, and the deviation data is compared with the data set I k Make updates;

[0183] The real-time observation module 400 is configured to use the offline calculated observation matrix M to perform Kalman filtering on the system and to obtain the augmented system state. and the error covariance Σ k Perform calculations;

[0184] The model predictive control module 500 is configured to approximate the infinite time domain optimal control problem based on the estimated system state and error covariance, and solve the finite time domain optimal control {u 0|k ,…,u N-1|k}, the controller output is used as the control input u of the real system k =u 0|k ;

[0185] The outer loop module 600 is configured to set k=k+1 and execute steps S20-S60 cyclically until the aircraft engine reaches the initially set system state; the prediction model and the spatiotemporal learning are constructed based on spatiotemporal Gaussian processes.

[0186] In order to verify the effectiveness of the designed control method in the aeroengine system, the present invention completes the system simulation verification of the aeroengine compressor control based on Matlab software. The simulation process is described as follows:

[0187] Schematic diagram of aircraft engine compressor system Figure 4 As shown, it is assumed that the distance between the coupling valve and the compressor outlet is close enough, and the storage of the gas mass flow rate is assumed to be negligible. For the compressor surge model, environmental factors such as wind speed, wind direction, air temperature, atmospheric pressure, intake conditions, etc. are considered as external environmental disturbances, which have an impact on the throttle opening, resulting in a decrease in control performance.

[0188] First consider the dynamic system of aircraft engine compressor control:

[0189]

[0190] Where φ represents the mass flow rate, ψ is the pressure rise, F is the control input force, ζ is the damping coefficient, ω n is the resonant frequency, β is a constant parameter, ψc is the pressure constant, r, They represent throttle opening, throttle opening rate, and throttle opening acceleration respectively. The system parameters are set as ψ c =0, β=1 and B d =5×10 -4 The state and input constraints are described as φ∈[0,1],ψ∈[1.1875,2.1875], r∈[0.1547,2.1547], And F∈[0.1547,2.1547].

[0191] In the numerical simulation, we set the steady-state φ ss =0.5,ψ ss =1.6875, r_{ss}=1.1547 and Using steady-state point linearization with sampling time Ts = 10 ms, the compressor system can be modeled as a discrete-time state-space equation, such as:

[0192] x k+1 =Ax k +Bu k +B d (g p,k +w k ) (32)

[0193] in u=F and

[0194] For the above compressor control system, find the optimal control input u with unknown potential * , so that the closed-loop system is stable at the steady-state point, and the cost function is as follows:

[0195]

[0196] We are in T s =0.01s for 10s time range for 20 instances under different operating conditions. In addition, we set the initial state to x0=[0.35,1.4875,1.1547,0] T , and set the state and control input cost matrices to

[0197] The aircraft engine system will be affected by some external environmental factors, such as wind speed, wind direction, temperature, atmospheric pressure, intake conditions, etc. Therefore, the spatiotemporal learning algorithm can be used to optimize the control of the compressor. The aircraft engine is in a stable state. We consider 1000 operating conditions within the flight envelope, including 800 points in known positions and 200 points in unknown positions. The 1000 numerical simulation positions on the flight envelope are as follows Figure 5 As shown, the Gaussian process model with the Matern covariance function is represented as a state-space model using the smoothing parameter

[0198] For comparison purpose, the traditional GP method and the proposed algorithm ST-MPC are applied.

[0199] The temporal evolution of the potential under different operating conditions is shown in Figure 6 As shown, Figure 6 (a) is an estimated comparison curve of external disturbance at a measurable node of an aircraft engine according to an embodiment of the present invention, and (b) is an estimated comparison curve of external disturbance at an unmeasurable node of an aircraft engine according to an embodiment of the present invention;

[0200] In the figure: True Latent Force represents the external disturbance of position; Conventional GP mean represents the mean of the traditional Gaussian process; Conventional GP covariance represents the variance of the traditional Gaussian process; Efficient Spatiotemporal learning mean represents the mean of spatiotemporal learning; Efficient Spatiotemporal learning covariance represents the variance of spatiotemporal learning; Figure 6 It can be seen that the covariance of the predictions at the virtual points is larger than the covariance of the estimates and measurements under operational conditions. It is worth noting that the proposed algorithm significantly reduces the computational requirements compared to the traditional space-time field methods.

[0201] By observation Figure 7 , Figure 7 (a) is a comparison curve of the mass flow rate of an aircraft engine under external disturbance of an embodiment of the present invention, and (b) is a comparison curve of the throat opening of an aircraft engine under external disturbance of an embodiment of the present invention; in the figure, Nominal MPC represents the standard model predictive control method; Proposed MPC represents the proposed model predictive control method; Reference represents the reference value, and Uncertainty represents the uncertainty. Figure 7 It can be found that the controller can stabilize each control variable at the set level in the presence of external disturbances.

[0202] As simulation results of the proposed algorithm, the state trajectories under different operating conditions in the identified compressor map are shown in Figure 8 As shown in the figure, Initial condition represents the initial condition; Desired equilibrium represents the designed equilibrium point; Nominal MPC represents the standard model predictive control method; Proposed MPC represents the standard model predictive control method, which are calculated based on the compressor characteristics. The surge control line is determined by a large number of physical experiments to describe the safe distance from the surge line. Figure 8 In , we note that the algorithm has satisfactory control performance under external environmental disturbances, which can be predicted more accurately by adopting spatiotemporal learning.

[0203] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A model predictive control method for an aircraft engine based on spatiotemporal learning, characterized in that: The following steps are involved: S10: Establishing the nonlinear dynamic equation of the aircraft engine and the real-time operating status of the aircraft engine; S20: Based on the nonlinear dynamic equations of the aircraft engine and the real-time operating state of the aircraft engine, a model predictive control algorithm based on spatiotemporal learning is used to predict the dynamic information of the actual aircraft engine operation process and improve the real-time performance of the control input solution; The process of using a model predictive control algorithm based on spatiotemporal learning based on the nonlinear dynamic equation of the aircraft engine and the real-time operation state of the aircraft engine to realize the prediction of dynamic information during the actual operation of the aircraft engine and improve the real-time performance of the control input solution is as follows: S201: Establishing an aircraft engine small deviation gain model with the current actual input of the aircraft engine as the steady-state point, after obtaining the linear expressions of the aircraft engine state variables and output variables, using the state space expression to construct the control model of the aircraft engine, and at the same time, given the initial state x0 of the system, constructing the state space expression of the space-time Gaussian process, offline calculating the observation matrix M and the feedback matrix K, and selecting the quadratic weights Q, R; S202: Get aircraft engine t k The actual state measurement value y at the moment k As well as the expected state variables, calculate the estimation error, and measure the external environment parameters p k ; S203: Calculate t k The deviation between the measured value of the aircraft engine system state and the nominal model at the moment, and the deviation data is compared with the data set I k Make updates; S204: Use the offline calculated measurement matrix M to perform Kalman filtering on the system and calculate the augmented system state and the error covariance Σ k Perform calculations; S205: Based on the estimated system state and error covariance, the infinite time domain optimal control problem is approximated to solve the finite time domain optimal control {u 0|k ,…,u N-1|k }, the output of the aircraft engine controller is used as the control input u of the real system k =u 0|k ; S206: Update the iteration sequence k=k+1, and return to S202 until the aircraft engine stops running.

2. The aircraft engine model predictive control method based on spatiotemporal learning according to claim 1, characterized in that: The control model of the aircraft engine is expressed as follows: In the formula, A, B, C are the discretized parameter matrices, k is the discrete time step, They are respectively represented as the system state, control input, measurement output and operating condition at the time instant, and the unknown dynamic g p,k It is expressed as the influence of external disturbance on the system at time k and hyperparameter p, w k is the process noise, v k To measure noise.

3. The aircraft engine model predictive control method based on spatiotemporal learning according to claim 2, characterized in that: The process of solving the expression of the control model of the aircraft engine is as follows: Assume a vector variable and a continuous mapping function A Gaussian process is described as: Where, μ(ξ), They are respectively expressed as the mean and covariance of the corresponding function value f(ξ). In order to reduce the problem of excessive computational requirements caused by long-term control, we first Perform Fourier transform to obtain the corresponding n-order Fourier domain stochastic differential equation: In the formula, is a constant coefficient, w(iω p ,t) The driving function is Gaussian white noise; By solving the inverse Fourier transform in the above equation, it is converted from the spatial Fourier domain to the spatial domain, and the corresponding state space expression is obtained using control theory: The state variables are The system matrix is ​​expressed as: Therefore, given the state-space equation, the power spectral density of g(p,t) is calculated as: where σ w is the power spectral density of Gaussian white noise, H(-iω) is the system transfer function, and the stable kernel function is obtained according to the inverse Fourier transform: By combining the mechanical model of the aircraft engine, an augmented state space model is obtained: In the formula, the system parameter matrix is:

4. The aircraft engine model predictive control method based on spatiotemporal learning according to claim 1, characterized in that: The process of improving the real-time performance of the control input solution is as follows: The controller parameterization is used to calculate the control input sequence. The asymptotically stable feedback matrix K and the mean square stable measurement matrix M are calculated offline to achieve the following controller parameterization: in Expressed as χ k A priori estimate of χ k The posterior estimate, based on the definition of the controller parameterization, obtains the corresponding predicted state at step i With control input u i|k : Therefore, the predicted state sequence and the estimated error series is represented as: in, Consider the estimation error σ k The covariance matrix of The update formula of the covariance matrix is: According to the above analysis, the corresponding predictive control input sequence u is obtained k , the estimated state sequence and the estimated error series δ k : in 5. The aircraft engine model predictive control method based on spatiotemporal learning according to claim 1, characterized in that: The aero-engine objective function is: The infinite time domain is reconstructed as: where the terminal term is defined as: Terminal term, make the following definition: Through the above definition, we get: The parameter matrix in the formula is described as According to P k The definition of , infers: Therefore, the terminal term c N (x i|N ) by calculating tr(ZP k ) is obtained, where At the same time k Represents the solution to the stochastic Lyapunov equation: