Direct Performance Adaptive Prediction Control Method and Device for Aeroengine

Abstract

The present invention discloses a direct performance adaptive prediction control method for an aeroengine. In the present invention, a tracking weight that is highly sensitive to the control effect and has no requirements for software / hardware is selected from the design parameters of the direct performance prediction controller as an adaptive adjustment parameter. Without adding additional design parameters, three variables, namely the PLA angle ratio, the relative tracking error, and the empirical proportional adjustment factor, are introduced to map the reasonable values of the tracking weight under the full envelope and full state. According to this mapping relationship, the tracking weight of the direct performance prediction controller is adaptively adjusted online. The present invention also discloses a direct performance adaptive prediction control device for an aeroengine. Compared with the existing direct performance prediction controller with constant design parameters, the present invention can achieve better control effects and dynamic response performances in the full life cycle, full envelope, and full state.

Description

Technical Field

[0001] The present invention relates to the technical field of aeroengine control, and particularly to a direct performance prediction control method for an aeroengine. Background Art

[0002] An aeroengine must work in a feedback control manner. The purpose of its control system is to achieve good thrust response ability and maintain important engine outputs within safety limits. However, with the high-performance technical requirements of new-generation aeroengines for future aircraft, the traditional indirect performance control method seems to be insufficient. Therefore, the currently developed advanced direct performance control technology has been widely favored, which can maximize the engine performance and achieve rapid thrust response. Although classical control methods or intelligent learning algorithms can be used for direct performance control currently, they still have shortcomings in constraint handling and performance optimization. Therefore, direct performance (thrust, surge margin, etc.) prediction control that is good at handling constraints and has optimization ability has become a research hotspot.

[0003] Direct performance predictive control mainly includes three parts: a prediction model, rolling optimization, and feedback correction. Since an aeroengine is a highly complex, strongly nonlinear, and strongly coupled time-varying system, the biggest challenge in implementing its direct performance predictive control is how to obtain a prediction model within the full state, full envelope, and full life cycle. According to existing research, currently, by linearizing the engine to establish piecewise linear models and linear parameter-varying models as prediction models, although the real-time performance is high, the accuracy is insufficient and the full envelope coverage rate is low. By establishing a simplified real-time nonlinear model as the prediction model, although it can follow the working state of the standard engine in real time when meeting the real-time requirements, the accuracy is also insufficient. In addition, using the state-space model obtained by real-time linearization of component-level models as the prediction model can follow and adapt to the working state of the engine within the full life cycle in real time, but it is difficult to meet the real-time requirements, and the accuracy of the real-time state-space model solved by the small perturbation method is not high. To solve the problems such as low accuracy caused by the weak adaptability of these model-based predictive control methods to the engine's nonlinearity, the engine nonlinear predictive control method based on intelligent learning has emerged. They use artificial intelligence learning algorithms to perform offline learning on the engine flight data to obtain a data-driven prediction model. However, the key to this method is that a large amount of flight test data is required, which is difficult to achieve for large envelope engines. But whether it is the direct performance predictive control designed by the prediction models obtained through the above-mentioned methods of establishing mathematical models or intelligent learning methods, there are still two common disadvantages in engineering applications: 1) It is difficult to adapt to the performance degradation of the engine throughout its life cycle, and the adaptive ability is weak; 2) The design parameters (prediction horizon, control horizon, control weight, and tracking weight) in the predictive control remain constant, and it can only achieve good control under local working conditions, while in some working conditions, there are phenomena such as overshoot and poor response. In a latest study, although a direct performance predictive control method for aeroengines based on subspace identification aims to solve the problem of difficult acquisition of the prediction model within the full state, full envelope, and full life cycle, and the obtained prediction model has high accuracy and the ability to adapt to the performance degradation of the engine throughout its life cycle, it still does not solve the shortcoming 2). Its constant design parameters still cannot ensure good control effects and dynamic response performance under the full life cycle, full envelope, and full state. For example, at a small throttle lever angle, there is a large overshoot and significant differences in control performance at different envelope points.

[0004] Therefore, there are still technical deficiencies in the existing direct performance predictive control of aeroengines, making it difficult to be truly applied. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to overcome the deficiencies brought about by the existing direct performance prediction control scheme for aero-engines using constant design parameters, and to provide a direct performance adaptive prediction control method for aero-engines, which adaptively adjusts the tracking weight online to ensure that the engine has good control effects and dynamic response performance throughout its entire life cycle, full flight envelope, and all operating states.

[0006] The present invention specifically adopts the following technical solutions to solve the above technical problems:

[0007] A direct performance adaptive prediction control method for an aero-engine, which performs direct performance prediction control on the aero-engine; the tracking weight Q used for the direct performance prediction control is dynamically adjusted according to three variables, namely the proportion β of the PLA angle, the relative tracking error δ, and the empirical proportional adjustment factor α. Among them, the proportion of the PLA angle t represents the proportion of the throttle lever angle PLA above idle compared to the maximum throttle lever angle PLA , and it has a non-linear function relationship proportional to the tracking weight Q max ; the relative tracking error t represents the relative tracking error of the thrust. When the engine is in a steady state, its value is 0. When the engine is in a dynamic state, its absolute value has a non-linear function relationship proportional to the tracking weight Q . Fn represents the actual thrust value, Fn t represents the current desired thrust value, and Fn cmd represents the previous desired thrust value; the empirical proportional adjustment factor α is a function of altitude and Mach number, and its value range is [0.1, 10]. cmd,p

[0008] One of the preferred solutions is to dynamically adjust the tracking weight Q t according to the following non-linear function relationship:

[0009]

[0010] In the formula, e is the natural constant.

[0011] Another preferred solution is to use a fuzzy logic regulator to dynamically adjust the tracking weight Q t . The fuzzy logic regulator takes β and δ as input variables, and Q t defined by the following formula is the output variable:

[0012] Q t = α·q t

[0013] where q t represents the output variable of the fuzzy logic, and its value range is [0.25, 1.1].

[0014] Further preferably, the fuzzy logic regulator uses Mamdani - type fuzzy logic.

[0015] Based on the same inventive concept, the following technical solutions can also be obtained:

[0016] An aero - engine direct performance adaptive prediction control device includes a direct performance prediction controller (SIIMPC) for directly predicting and controlling the performance of an aero - engine; the device further includes an adaptive tracking weight module for dynamically adjusting the tracking weight Q of the direct performance prediction controller according to three variables: the proportion β of the PLA angle, the relative tracking error δ, and the empirical proportional adjustment factor α. Among them, the proportion of the PLA angle t represents the proportion of the throttle lever angle PLA above the idle speed compared to the maximum throttle lever angle PLA and has a non - linear function relationship proportional to the tracking weight Q max ; the relative tracking error t represents the relative tracking error of the thrust. When the engine is in a steady state, its value is 0. When the engine is in a dynamic state, its absolute value has a non - linear function relationship proportional to the tracking weight Q . Here, Fn represents the actual thrust value, Fn t represents the current desired thrust value, and Fn cmd represents the previous desired thrust value; the empirical proportional adjustment factor α is a function of altitude and Mach number, and its value range is [0.1, 10]. cmd,p

[0017] One of the preferred solutions is that the adaptive tracking weight module dynamically adjusts the tracking weight Q according to the following non - linear function relation: t

[0018]

[0019] In the formula, e is the natural constant.

[0020] Another preferred solution is that the adaptive tracking weight module uses a fuzzy logic regulator to dynamically adjust the tracking weight Q. The fuzzy logic regulator takes β and δ as input variables, and Q t defined by the following formula is the output variable: t

[0021] Q t =α·q t

[0022] where q t represents the output variable of the fuzzy logic, and its value range is [0.25, 1.1].

[0023] ​​​Further preferably, the fuzzy logic regulator uses Mamdani-type fuzzy logic.

[0024] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0025] The present invention selects the tracking weight that is highly sensitive to the control effect from the design parameters of the direct performance prediction controller (SIIMPC) as the adaptive adjustment parameter. Without adding additional design parameters, three variables, namely the PLA angle ratio, the relative tracking error, and the empirical proportional adjustment factor, are introduced to map the reasonable values of the tracking weight under the full envelope and full state. According to this mapping relationship, the tracking weight of the SIIMPC is adaptively adjusted online. Compared with the existing SIIMPC with constant design parameters, better control effects and dynamic response performances can be achieved throughout the entire life cycle, full envelope, and full state. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a schematic diagram of the direct performance adaptive prediction control device for an aeroengine of the present invention;

[0027] Figure 2 It is a calculation flowchart defined for the variable PLA angle ratio β and the relative tracking error δ;

[0028] Figure 3 It is a structure diagram of the fuzzy logic regulator;

[0029] Figure 4 It is a membership function curve diagram of the input variable β in the fuzzy logic regulator;

[0030] Figure 5 It is a membership function curve diagram of the input variable δ in the fuzzy logic regulator;

[0031] Figure 6 It is for the output variable q t of the membership function curve diagram;

[0032] Figure 7 It is a decision diagram of the inference rule in the fuzzy logic regulator;

[0033] Figure 8 It is a three-dimensional diagram of the fuzzy logic regulator. DETAILED DESCRIPTION OF THE INVENTION

[0034] Aiming at the deficiencies brought about by the constant design parameters in the existing direct performance prediction control scheme for aero-engines, the solution idea of the present invention is to select the tracking weight, which is sensitive to the control effect and has no requirements for software / hardware, from the design parameters of the direct performance prediction controller (SIIMPC) as the adaptive adjustment parameter. Without adding extra design parameters, three variables, namely the PLA angle ratio, the relative tracking error, and the empirical proportional adjustment factor, are introduced to map the reasonable value of the tracking weight under the full envelope and full state. And the tracking weight of the SIIMPC is adaptively adjusted online according to this mapping relationship.

[0035] The present invention specifically adopts the following technical solutions to solve the above technical problems:

[0036] A direct performance adaptive prediction control method for aero-engines, which performs direct performance prediction control on the aero-engine; the tracking weight Q used for the direct performance prediction control is dynamically adjusted according to three variables, namely the PLA angle ratio β, the relative tracking error δ, and the empirical proportional adjustment factor α. Among them, the PLA angle ratio t represents the ratio of the throttle lever angle PLA above the idle to the maximum throttle lever angle PLA and has a non-linear function relationship proportional to the tracking weight Q. The relative tracking error max represents the relative tracking error of the thrust. When the engine is in a steady state, its value is 0. When the engine is in a dynamic state, its absolute value has a non-linear function relationship proportional to the tracking weight Q. Fn represents the actual thrust value, Fn t represents the current desired thrust value, and Fn represents the previous desired thrust value. The empirical proportional adjustment factor α is a function of altitude and Mach number, and its value range is [0.1, 10]. t cmd cmd,p

[0037] A direct performance adaptive prediction control device for aero-engines, including a direct performance prediction controller (SIIMPC) for performing direct performance prediction control on the aero-engine; the device also includes an adaptive tracking weight module for dynamically adjusting the tracking weight Q of the direct performance prediction controller according to three variables, namely the PLA angle ratio β, the relative tracking error δ, and the empirical proportional adjustment factor α. Among them, the PLA angle ratio t represents the ratio of the throttle lever angle PLA above the idle to the maximum throttle lever angle PLA and has a non-linear function relationship proportional to the tracking weight Q. The relative tracking error max represents the ratio of the throttle lever angle PLA above the idle to the maximum throttle lever angle PLA t and has a non-linear function relationship proportional to the tracking weight Q. The relative tracking error ​​​Represents the relative tracking error of thrust. When the engine is in a steady state, its value is 0. When the engine is in a dynamic state, its absolute value is a non-linear function relationship proportional to the tracking weight Q t and is a non-linear function relationship proportional to the tracking weight Q. Fn represents the actual thrust value, Fn cmd represents the current desired thrust value, Fn cmd,p represents the previous desired thrust value; the empirical proportional adjustment factor α is a function of altitude and Mach number, and its value range is [0.1, 10].

[0038] For the convenience of public understanding, the technical solution of the present invention will be described in detail below through a specific embodiment in combination with the accompanying drawings:

[0039] The basic structural principle of the aero-engine direct performance adaptive prediction control device in this embodiment is as Figure 1 shown, including an SIIMPC controller and an adaptive tracking weight module. The SIIMPC controller is used for direct performance prediction control of the aero-engine, and the adaptive tracking weight module is used to dynamically adjust the tracking weight Q of the direct performance prediction controller according to the three variables of the PLA angle ratio β, the relative tracking error δ, and the empirical proportional adjustment factor α t for dynamic adjustment.

[0040] (1) Construct the following SIIMPC controller based on subspace identification:

[0041] 1-1) Continuously apply the persistent excitation input signal u k (k = 1, 2,..., N) to the system near the engine idle point, and construct the past / future control input variable full row rank Hankel matrix (U ct ) and the past / future output variable full row rank Hankel matrix (Y p / U f ) and the past / future output variable full row rank Hankel matrix (Y ct,p / Y ct,f ) in sequence from the obtained input / output (I / O) data set D(U, Y N ). In particular, U = [u1, u2,..., u N , u is a control input variable with a dimension of nu; Y ct = [y ct,1 , y ct,2 ,..., y ct,N , y ct represents the constrained output variable (y c ) with a dimension of ny c and the tracking output variable (y t ) with a dimension of ny t combined into a data set with a dimension of ny ct , that is, y ct = [y c , yt T ;

[0042] 1 - 2) Design the future optimal prediction output estimator in the form of

[0043]

[0044] wherein, L w and L u respectively represent the subspace prediction estimator coefficient matrices of past I / O data and future input data. W p is the combination of past input / output Hankel matrices, defined as

[0045] 1 - 3) Use the LQ decomposition method to decompose the following combined Hankel matrix

[0046]

[0047] wherein, is a lower triangular matrix, R 11 , R 21 , R 22 , R 31 , R 32 and R 33 are block matrices of the R matrix, is an orthogonal matrix, and Q1, Q2 and Q3 are block matrices of Q.

[0048] Then, the values of L w and L u are

[0049]

[0050] where the superscript represents the generalized inverse.

[0051] 1 - 4) Obtain the future f - step prediction output equation as

[0052]

[0053] wherein, p is the past time domain, f is the prediction time domain, satisfying f = p, m is the control time domain, is the predicted value of y ct .

[0054] 1 - 5) Deduce the future f - step incremental prediction output equation at time k

[0055]

[0056] ​Among them, y ct,k is the combined vector of the constrained output and the tracking output at time k, I is the identity matrix.

[0057] According to the constrained output variable y c and the tracking output variable y t split the incremental predictive output equation to obtain

[0058]

[0059]

[0060] In the formula, represents the constrained output of the future f-step prediction, F 1,c , F 2,c , Δw p,c and respectively represent their corresponding coefficient matrices, y c,k is the constrained output at time k; represents the tracking output of the future f-step prediction, F 1,t , F 2,t , Δw p,t and respectively represent their corresponding coefficient matrices, y t,k is the tracking output at time k.

[0061] 1 - 6) In the discrete-time system of the engine, the finite-horizon quadratic cost function aiming to achieve optimal output tracking and minimum fuel consumption has the following form:

[0062]

[0063]

[0064] Among them,

[0065] The non-bold Q t is the positive tracking weight; the non-bold R u is the positive control weight, r is the desired engine command, i.e., the thrust Fn, u min and u max respectively represent the minimum and maximum values restricted by the control quantity u k . y c,min and y c,max represent the minimum and maximum values restricted by the constrained output.

[0066] (2) Online Adaptive Adjustment Method for Tracking Weight Based on Multivariable Mapping

[0067] 2-1) Variable Definition

[0068] Based on the adaptive predictive control method in (1), without additional design parameters, considering that the prediction horizon f and the control horizon m affect the dimensions of the relevant matrices in predictive control. Changing them will seriously affect the complexity of the control system code and dynamic memory allocation, and have strict requirements for software / hardware. Therefore, it is not conducive to adjusting them to ensure the performance of the controller. For the control weight R u However, its order of magnitude is often large, and the sensitivity of numerical change to control performance is weak. On the contrary, the tracking weight Q t has a small value and is more sensitive to control performance. Adjusting it is the best choice. Therefore, the following three variables are defined to characterize the non-linear adjustment law of the tracking weight Q t to achieve the optimal adjustment of the direct performance predictive controller:

[0069] α: It is an empirical proportional adjustment factor, which is only a function of altitude and Mach number, that is, different altitudes and Mach numbers correspond to different values. When the altitude and Mach number are fixed, it does not change with the engine operating state. In the designed predictive control, the general value range is [0.1, 10].

[0070] β: It represents the ratio of the throttle lever angle (PLA) above idle to the maximum throttle lever angle (PLAmax), that is, β ∈ [0.5, 1.0]. The tracking weight Q t has a non-linear functional relationship proportional to β. The larger β is, the larger Q t is; the smaller β is, the smaller Q t is. It can effectively solve the overshoot phenomenon when the value of Q t remains unchanged under small PLA values. It is defined as follows:

[0071]

[0072] where PLA ∈ [40°, 80°] is the throttle lever angle, and PLA max is the maximum throttle lever angle, with a value of 80°.

[0073] δ: It represents the relative tracking error of the thrust (Fn), and the value range is [-1, 1]. At the engine steady state, the value is 0. During engine dynamics, Q t has a non-linear functional relationship proportional to the absolute value of δ. The larger the absolute value of δ is, the larger Q t is; the smaller the absolute value of δ is, the smaller Q t is. It can effectively improve the dynamic acceleration / deceleration performance during the predictive control process, such asFigure 2 As shown. The definition is as follows:

[0074]

[0075] Among them, Fn cmd Indicates the current expected thrust value, Fn cmd,p Indicates the previous expected thrust value.

[0076] According to the above rules, establish Q t The nonlinear mapping function relationship between α, β and δ can realize Q t Adaptive adjustment in the process of direct engine performance prediction control further improves the control performance. The present invention will provide two verified and satisfactory nonlinear mapping relationship design methods:

[0077] 2-2) Online adaptive adjustment method of tracking weight

[0078] Method 1:

[0079] Establish the Q shown below t The nonlinear functional relationship with α, β and δ can quickly realize Q t Adaptive adjustment according to engine status changes:

[0080]

[0081] Among them, in the second Middle, Q t It is a nonlinear polynomial function proportional to β. In , min(·) represents the minimum value operator, such as Indicated in and 1, whichever is smaller. The term indicates that when β is constant, the effect of δ on Q t When δ>0, The term is proportional to δ, and when -0.5<δ<0, The term is inversely proportional to δ; when -1.0<δ<-0.5, When PLA decreases (deceleration), β decreases, Q t Decreasing will slow down the deceleration process. At this time, δ is a negative value. The project will play a positive role in alleviating Q t The decrease of the value speeds up the deceleration process. When PLA becomes larger (acceleration), β becomes larger, Q t becomes larger. At this time, δ is a positive value, The term will play a positive role, |δ| stimulates Q tThe value further increases, enhancing the control effect and accelerating the acceleration process.

[0082] Method 2:

[0083] According to the non - linear mapping function relationship between Q t and α, β, and δ, with β and δ as input variables, design a fuzzy logic regulator within the full - envelope of the engine with Q defined by the following formula t as the output variable, as shown in Figure 3 It adopts Mamdani - type fuzzy logic with a typical "IF - THEN" rule structure.

[0084] Q t = α·q t

[0085] where q t represents the output variable of the fuzzy logic, and the empirical value range is set to [0.25, 1.1].

[0086] Specifically, fuzzy logic is divided into three parts: fuzzification, inference rules, and defuzzification. Among them, in fuzzification, as shown in Figure 4 , the first input variable β is represented as a linguistic fuzzy variable by a fuzzy set established by 10 different membership functions, where all are triangular except the right - most one which is trapezoidal: Very Small (SV), Pretty Small (SP), Small (SS), Quite Small (SQ), General Small (SG), Medium (MM), General Big (BG), Quite Big (BQ), Big (BB), and Pretty Big (BP). As shown in Figure 5 , the second input variable δ is represented as a linguistic fuzzy variable by a fuzzy set established by triangular - shaped membership functions: Negative Big (NB), Negative Medium (NM), Negative Small (NS), Zero (ZZ), Positive Small (PS), Positive Medium (PM), and Positive Big (PB). In the inference rules, as shown in Figure 6 , the output variable q of the fuzzy logic tThe value range design represents the linguistic fuzzy variables formed by the fuzzy sets established by the membership functions in the shape of triangles as: Very Small (SV), Pretty Small (SP), Small (SS), Quite Small (SQ), General Small (SG), Medium (MM), General Big (BG), Big (BB), Pretty Big (BP), and Very Big (BV). The larger β and δ are, t the larger q is, and vice versa. Therefore, the rules designed according to this reasoning law are determined as Figure 7 shown. In defuzzification, the general center of area method is selected to convert the fuzzy quantity obtained by reasoning into an accurate quantity. Finally, the accurate quantity q t obtained by defuzzification is multiplied by the proportional adjustment factor α to obtain the fuzzy logic regulator as Figure 8 shown.

Claims

1. A direct performance adaptive prediction control method for an aeroengine, which performs direct performance prediction control on the aeroengine; characterized in that, Adjust the tracking weight Q used in the direct performance predictive control according to three variables: the PLA angle ratio β, the relative tracking error δ, and the empirical proportional adjustment factor α t dynamically, where the PLA angle ratio represents the ratio of the throttle lever angle PLA above the idle to the maximum throttle lever angle PLA max and has a non-linear functional relationship proportional to the tracking weight Q t ; the relative tracking error represents the relative tracking error of the thrust. When the engine is in a steady state, its value is 0. When the engine is in a dynamic state, its absolute value has a non-linear functional relationship proportional to the tracking weight Q t , Fn represents the actual thrust value, Fn cmd represents the current desired thrust value, Fn cmd,p represents the previous desired thrust value; the empirical proportional adjustment factor α is a function of altitude and Mach number, and its value range is [0.1, 10].

2. The direct performance adaptive prediction control method for an aeroengine according to claim 1, wherein The tracking weight Q is dynamically adjusted according to the following non-linear functional relationship: t For dynamic adjustment: In the formula, e is the natural constant.

3. The direct performance adaptive predictive control method for an aero-engine according to claim 1, wherein Dynamically adjust the tracking weight Q using a fuzzy logic regulator t The fuzzy logic regulator takes β and δ as input variables, and Q defined by the following formula t is the output variable: Q t = α·q t where q t represents the output variable of fuzzy logic, and its value range is [0.25, 1.1].

4. The direct performance adaptive prediction control method for an aeroengine according to claim 3, characterized in that, The described fuzzy logic regulator uses Mamdani-type fuzzy logic.

5. An aircraft engine direct performance adaptive prediction control device, including a direct performance prediction controller for directly performing performance prediction control on an aircraft engine; characterized in that, The device further includes an adaptive tracking weight module for dynamically adjusting the tracking weight Q of the direct performance prediction controller according to three variables: the PLA angle ratio β, the relative tracking error δ, and the empirical proportional adjustment factor α. t Among them, the PLA angle ratio represents the ratio of the throttle lever angle PLA above the slow speed to the maximum throttle lever angle PLA max , and it has a non-linear function relationship proportional to the tracking weight Q t ; the relative tracking error represents the relative tracking error of the thrust. When the engine is in a steady state, its value is 0. When the engine is in a dynamic state, its absolute value has a non-linear function relationship proportional to the tracking weight Q t . Fn represents the actual thrust value, and Fn cmd represents the current desired thrust value, and Fn cmd,p represents the previous desired thrust value; the empirical proportional adjustment factor α is a function of altitude and Mach number, and its value range is [0.1, 10].

6. The aeroengine direct performance adaptive prediction control device according to claim 5, characterized in that The adaptive tracking weight module dynamically adjusts the tracking weight Q according to the following non-linear function relationship t as follows: In the formula, e is the natural constant.

7. The direct performance adaptive prediction control device for an aeroengine according to claim 5, characterized in that, The adaptive tracking weight module uses a fuzzy logic regulator to dynamically adjust the tracking weight Q t where the fuzzy logic regulator takes β and δ as input variables and Q defined by the following equation t as the output variable: Q t = α·q t where q t represents the output variable of fuzzy logic, and its value range is [0.25, 1.1].

8. The aeroengine direct performance adaptive prediction control device according to claim 7, characterized in that, The described fuzzy logic regulator uses Mamdani-type fuzzy logic.

Images