Data-driven multivariable adaptive predictive control method for adaptive cycle engines

Abstract

This application provides a data-driven multivariable adaptive predictive control method for an adaptive cyclic engine, belonging to the field of aero-engine technology. The method includes: constructing an equivalent data model of the adaptive cyclic engine based on full-format dynamic linearization of the engine's input and output data to obtain multi-step forward output prediction equations; performing real-time estimation of the pseudo-gradient matrix in the prediction equations using autoregressive and projection algorithms to form a predictive control rolling optimization framework; embedding a model-free adaptive control algorithm based on single-step optimization into this framework to form a composite control strategy; designing control input performance indicators with the goal of minimizing tracking error to obtain optimal control parameters; and generating real-time optimal control parameters using the engine's real-time input and output data based on the composite control strategy to achieve precise multivariable coordinated control. This application improves the long-term prediction, constraint handling, and global optimization capabilities of the control system.

Description

Technical Field

[0001] This application relates to the field of aero-engine technology, and in particular to a data-driven adaptive cyclic engine multivariable adaptive predictive control method. Background Technology

[0002] The adaptive cyclic engine achieves wide-range bypass ratio adjustment by introducing a third bypass flow path and a core engine-driven fan stage, and possesses the ability to switch between multiple operating modes. Its dynamic process exhibits strong nonlinearity, multivariable coupling, and time-varying parameters. Traditional predictive control performance heavily relies on an accurate mathematical model of the controlled object. However, the adaptive cyclic engine has complex dynamic characteristics, variable operating states, numerous adjustable parameters, and strong coupling between internal parameter variables, making it difficult to establish an accurate mathematical model, thus leading to a decline in predictive control performance. While model-free adaptive control does not require prior model information, it suffers from slow response speed and large overshoot. Summary of the Invention

[0003] In view of this, embodiments of this application provide a data-driven multivariable adaptive predictive control method for adaptive cycle engines, which at least partially solves the control problems faced by adaptive cycle engines under complex operating conditions, such as strong nonlinearity, model uncertainty, and multivariable coupling. This application employs a model-free adaptive predictive control method, relying solely on real-time engine input and output data to establish a full-format dynamic linearized model online and estimate the pseudo-gradient matrix in real time. Combined with a rolling optimization strategy, optimal control parameters are dynamically generated, ultimately achieving precise coordinated control of multivariables at high-pressure speed, pressure ratio, and low-pressure speed. This significantly improves the engine's control accuracy, dynamic adaptability, and robustness across the entire control envelope.

[0004] This application provides a data-driven adaptive cyclic engine multivariable adaptive predictive control method, the method comprising:

[0005] Step 1: Based on the input and output data of the adaptive cycle engine, construct an equivalent data model of the adaptive cycle engine based on full-format dynamic linearization, and obtain the multi-step forward output prediction equation based on the equivalent data model.

[0006] Step 2: Based on the autoregressive and projection algorithms, the pseudo gradient matrix in the multi-step forward output prediction equation is estimated online in real time to form a rolling optimization framework for predictive control.

[0007] Step 3: Embed the model-free adaptive control algorithm based on single-step optimization into the rolling optimization framework to form a composite control strategy. Design the control input performance index with the goal of minimizing the tracking error and obtain the optimal control parameters.

[0008] Step 4: Based on the composite control strategy, the optimal control parameters are calculated and generated online using the real-time input and output data of the adaptive cycle engine, thereby achieving precise and coordinated multi-variable control of the engine.

[0009] According to a specific implementation of an embodiment of this application, the construction of an equivalent data model of an adaptive cyclic engine based on full-format dynamic linearization includes:

[0010] Step 11: Based on the analysis of the mapping relationship between the opening degree of the variable geometry components of the adaptive cycle engine and the overall engine performance, determine the control variables and controlled parameters, and construct a multivariable control framework reflecting the internal energy conversion and thrust output of the engine. The expression is:

[0011] ,

[0012] Where u(k) is the input of the system at time k, y(k) is the output of the system at time k, and n u Input an unknown order, n, into the system. y The system output has an unknown order, and f(·) is a nonlinear function;

[0013] Step 12, Definition To input the relevant sliding time window All control input signals within and in the output-related sliding time window The first vector is composed of all system output signals within it. The expression is:

[0014] ,

[0015] ,

[0016] When k≤0 is satisfied ,

[0017] Where R is the set of real numbers, and m is the dimension. Let be the first pseudo-order of the system. The second pseudo-order of the system, ;

[0018] Step 13: When the expression of the multivariate control framework satisfies conditions one and two, then when At that time, there exists a pseudo-gradient time-varying parameter that transforms the first vector into an equivalent data model.

[0019] The first condition is: the nonlinear function f of the object i (·) has continuous partial derivatives with respect to the components of each variable, where i is the first variable, i=1,2,…,m;

[0020] The second condition is: for any two times k1 and k2... , ,have ω is the first constant greater than 0. This represents the change in the first vector at time k. This represents the change in the system's output at time k+1.

[0021] According to a specific implementation of an embodiment of this application, the controlled quantities include the main fuel flow rate, the tailpipe throat area, and the area of ​​the rear adjustable duct ejector, and the controlled parameters include high-pressure speed, low-pressure speed, and pressure ratio.

[0022] According to a specific implementation of this application, the expression for the pseudo-gradient time-varying parameter is:

[0023] , i=1,2,…,L y +L u ,

[0024] ,

[0025] in, Let be the pseudo-gradient time-varying parameters at time k. Let be the variable representing the pseudo-gradient time-varying parameter at time k. Let this be the first parameter variable at time k;

[0026] The expression for the equivalent data model is:

[0027] ,

[0028] It is bounded at any time k.

[0029] According to a specific implementation of an embodiment of this application, obtaining the multi-step forward output prediction equation based on an equivalent data model includes:

[0030] make: ,

[0031] ,

[0032] ,

[0033] ,

[0034] , , , ,

[0035] in, Let this be the first intermediate variable at time k. Let k be the change in the system's output at time k. Let k be the change in the system output at time k-1. The second intermediate variable at time k. Let be the change in the system's input at time k. Let be the change in the system's input at time k-1. Let be the third intermediate variable at time k. Let A be the fourth intermediate variable at time k, and let B be the first matrix, C be the second matrix, and D be the fourth matrix.

[0036] Set the prediction step size N=5, and set the control time domain constant N. u =2;

[0037] make: ,

[0038] ,

[0039] ,

[0040] ,

[0041] ,

[0042] ,

[0043] ,

[0044] The expression for the multi-step forward output prediction equation is:

[0045] ,

[0046] in, To predict output parameters, This is the fifth intermediate variable at time k. This is the fifth intermediate variable at time k+1. This is the sixth intermediate variable at time k. Let be the difference between the system input at time k+1 and the system input at time k. This is the first pseudo gradient matrix. This is the second pseudo-gradient matrix. This is the third pseudo-gradient matrix. This is the fourth pseudo-gradient matrix.

[0047] According to a specific implementation of an embodiment of this application, the online real-time estimation of the pseudo-gradient matrix in the multi-step forward output prediction equation based on autoregression and projection algorithms includes:

[0048] For the first pseudo-gradient matrix The second pseudo gradient matrix The third pseudo gradient matrix and the fourth pseudo gradient matrix The third intermediate variable at time k in and the fourth intermediate variable at time k The estimation is performed using a projection algorithm;

[0049] For the first pseudo-gradient matrix The second pseudo gradient matrix The third pseudo gradient matrix and the fourth pseudo gradient matrix The third intermediate variable at time k+i and the fourth intermediate variable at time k+i For i=1,2,3,4, a multi-level hierarchical prediction method is used for prediction.

[0050] According to a specific implementation of an embodiment of this application, the expression for estimation using a projection algorithm is as follows:

[0051] ,

[0052] in, Let μ be the change in the estimated value of the pseudo-gradient time-varying parameters, μ be the first step length factor, and η be the second step length factor, where μ > 0 and η ∈ (0, 1]. This represents the estimated pseudo-gradient time-varying parameter value at time k-1.

[0053] According to a specific implementation of an embodiment of this application, the expression of the autoregressive model of the multi-level hierarchical prediction method is as follows:

[0054] ,

[0055] ,

[0056] ,

[0057] ,

[0058] in, The pseudo-gradient time-varying parameter estimate at time k is... For an unknown matrix, For the nth p There are n unknown matrix variables.p For order, This is an intermediate variable for estimating time-varying parameters of the pseudo-gradient;

[0059] Unknown matrix The formula for the least squares algorithm with a forgetting factor is determined by the following expression:

[0060] ,

[0061] ,

[0062] ,

[0063] Where P(·) is the seventh intermediate variable, α(·) is the eighth intermediate variable, α0 is the second constant, and α0=0.99; when k=1, P(k-2)=P(-1)>0 and α(0)=0.95 are defined.

[0064] According to a specific implementation of an embodiment of this application, the functional expression of the control input performance index is:

[0065] ,

[0066] Where J is the control input performance index, and λ is the weighting factor. This is the desired output signal.

[0067] According to a specific implementation of an embodiment of this application, obtaining the optimal control parameters includes:

[0068] Based on the control input performance index, the multi-step forward output prediction equation is substituted and the results are analyzed. Taking the derivative and setting its first derivative to zero, we obtain the following expression:

[0069] ,

[0070] The optimal control parameters are obtained by employing a rolling optimization strategy, expressed as follows:

[0071] ,

[0072] ,

[0073] Where I is the identity matrix and g is the fifth matrix.

[0074] Beneficial effects:

[0075] The data-driven adaptive cyclic engine multivariable adaptive predictive control (MFAPC) method in this application is applicable to multivariable, strongly nonlinear aero-engine control systems. This method relies solely on real-time system input and output data, constructing an equivalent model online through dynamic linearization. Combined with pseudo-gradient matrix estimation and rolling optimization, it achieves high-precision coordinated control of multiple variables such as high-pressure speed, pressure ratio, and low-pressure speed, effectively overcoming control challenges such as system nonlinearity and strong coupling.

[0076] By deeply integrating predictive control with model-free adaptive control, the problem of dependence on accurate models in traditional prediction methods is effectively solved. Online adaptive linearization is used to provide a model basis for prediction methods, and the rolling optimization framework is used to improve the long-term prediction, constraint handling and global optimization capabilities of the control system, thereby reducing its dependence on prior knowledge. Attached Figure Description

[0077] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0078] Figure 1 This is a cross-sectional view of an adaptive cycle engine according to an embodiment of the present invention.

[0079] Figure 2 This is a schematic diagram of a data-driven adaptive cyclic engine multivariable adaptive predictive control method according to an embodiment of the present invention.

[0080] Figure 3a The following is a schematic diagram of the simulation results of the controlled parameters of the control method of the present invention applied in the three-function mode of the adaptive cycle engine: (1) is the simulation result of high pressure speed, (2) is the simulation result of low pressure speed, and (3) is the simulation result of pressure ratio.

[0081] Figure 3b The following is a schematic diagram of the control quantity simulation results of the control method of the present invention applied in the three-channel mode of the adaptive cycle engine: (1) is the simulation result of the main fuel flow rate, (2) is the simulation result of the throat area of ​​the tail nozzle, and (3) is the simulation result of the area of ​​the rear adjustable duct ejector.

[0082] Figure 4a The following is a schematic diagram of the simulation results of the controlled parameters of the control method of the present invention applied in the dual-function mode of the adaptive cycle engine: (1) is the simulation result of high pressure speed, (2) is the simulation result of low pressure speed, and (3) is the simulation result of pressure ratio.

[0083] Figure 4b The following is a schematic diagram of the simulation results of the control quantity of the control method of the present invention applied in the dual-bypass mode of the adaptive cycle engine: (1) is the simulation result of the main fuel flow rate, (2) is the simulation result of the throat area of ​​the tail nozzle, and (3) is the simulation result of the area of ​​the rear adjustable duct ejector. Detailed Implementation

[0084] The embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0085] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. This application can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0086] It should be noted that various aspects of embodiments within the scope of the appended claims are described below. It will be apparent that the aspects described herein can be embodied in a wide variety of forms, and any particular structure and / or function described herein is merely illustrative. Based on this application, those skilled in the art will understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects set forth herein can be used to implement the device and / or practice the method. Additionally, this device and / or method can be implemented using structures and / or functionalities other than one or more of the aspects set forth herein.

[0087] It should also be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of this application. The illustrations only show the components related to this application and are not drawn according to the number, shape and size of the components in actual implementation. In actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0088] Furthermore, specific details are provided in the following description to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that the described aspects can be practiced without these specific details.

[0089] This application provides a data-driven adaptive cyclic engine multivariable adaptive predictive control method, the method comprising:

[0090] Step 1: Based on the input and output data of the adaptive cycle engine, construct an equivalent data model of the adaptive cycle engine based on full-format dynamic linearization, and obtain the multi-step forward output prediction equation based on the equivalent data model.

[0091] Step 2: Based on the autoregressive and projection algorithms, the pseudo gradient matrix in the multi-step forward output prediction equation is estimated online in real time to form a rolling optimization framework for predictive control.

[0092] Step 3: Embed the model-free adaptive control algorithm based on single-step optimization into the rolling optimization framework to form a composite control strategy. Design the control input performance index with the goal of minimizing the tracking error and obtain the optimal control parameters.

[0093] Step 4: Based on the composite control strategy, the optimal control parameters are calculated and generated online using the real-time input and output data of the adaptive cycle engine, thereby achieving precise and coordinated multi-variable control of the engine.

[0094] In this embodiment, based on the input and output data of the adaptive cyclic engine, a full-format data model and multi-step output prediction are established online using a dynamic linearization method. A multivariable model-free adaptive predictive control strategy is proposed, which estimates the pseudo-gradient matrix online through autoregression and projection algorithms, and generates optimal control parameters in real time by combining rolling optimization, thereby improving the control performance of complex nonlinear systems. Using this control algorithm, the optimal control parameters are calculated and generated online based on the input and output data of the adaptive cyclic engine, realizing multivariable precise coordinated control of the engine's high-pressure speed, pressure ratio, and low-pressure speed.

[0095] This method relies solely on real-time input and output data of the system to construct an equivalent model online through dynamic linearization. Combined with pseudo-gradient matrix estimation and rolling optimization, it achieves high-precision coordinated control of multiple variables, including high-pressure speed, pressure ratio, and low-pressure speed of the engine, effectively overcoming control challenges such as system nonlinearity and strong coupling. By deeply integrating predictive control with model-free adaptive control, it effectively solves the problem of traditional predictive methods' dependence on accurate models. Online adaptive linearization provides a model foundation for the predictive method, and the rolling optimization framework enhances the long-term prediction, constraint handling, and global optimization capabilities of the control system, reducing its dependence on prior knowledge.

[0096] In specific implementation, the working section diagram of a certain type of adaptive cycle engine used is as follows: Figure 1As shown, its main components include the intake, FLADE, fan, core driven fan stage (CDFS), high-pressure compressor, main combustion chamber, high-pressure turbine, low-pressure turbine, afterburner, and exhaust nozzle. The nonlinear mathematical model of the engine is obtained by using the component method based on C language, and then encapsulated into a dynamic link library, which is called in the MALTAB environment for digital simulation verification.

[0097] The following example uses the steady-state control of a certain type of adaptive cycle engine. Figure 2 This is a structural diagram of a data-driven adaptive cyclic engine multivariable adaptive predictive control method. Figure 2 In the middle, Δn Hr The reference input is the change in high-pressure speed, Δn Lr ΔEPR is the reference input low-pressure speed change. r ΔW is the reference input pressure ratio change. fb The change in main fuel flow rate, ΔA8 is the change in nozzle throat area, ΔA RVABI Δn represents the change in the area of ​​the adjustable duct ejector. H Δn represents the change in high-pressure rotational speed. L ΔEPR represents the change in speed at low pressure, while ΔEPR represents the change in pressure ratio.

[0098] Furthermore, the construction of the equivalent data model of the adaptive cyclic engine based on full-format dynamic linearization includes:

[0099] Step 11: Based on the analysis of the mapping relationship between the opening degree of the variable geometry components of the adaptive cycle engine and the overall engine performance, determine the control variables and controlled parameters, and construct a multivariable control framework reflecting the internal energy conversion and thrust output of the engine. The expression is:

[0100] (1),

[0101] Where u(k) is the input of the system at time k, y(k) is the output of the system at time k, and n u Input an unknown order, n, into the system. y The system output has an unknown order, f(·) is a nonlinear function, and this embodiment uses a three-variable controller with m=3;

[0102] Step 12, Definition To input the relevant sliding time window All control input signals within and in the output-related sliding time window The first vector is composed of all system output signals within it. The expression is:

[0103] (2),

[0104] ,

[0105] When k≤0 is satisfied ,

[0106] Where R is the set of real numbers, and m is the dimension. Let be the first pseudo-order of the system. The second pseudo-order of the system, , and All numbers are integers; L is selected considering adaptive engine characteristics. y =L u =2, therefore we can get ;

[0107] Step 13: When the expression of the multivariate control framework satisfies conditions one and two, then when At that time, there exists a time-varying parameter called pseudo gradient (PG) that transforms the first vector into an equivalent data model (FFDL data model, Full-Form Dynamic Linearization).

[0108] The first condition is: the nonlinear function f of the object i (·) has continuous partial derivatives with respect to the components of each variable, where i is the first variable, i=1,2,…,m;

[0109] The second condition is: the Lipschitz condition must be satisfied, that is, for any two times k1 and k2, , ,have ω is the first constant greater than 0. This represents the change in the first vector at time k. This represents the change in the system's output at time k+1.

[0110] Furthermore, the controlled quantities include the main fuel flow rate, the tailpipe throat area, and the rear adjustable duct ejector area, and the controlled parameters include high-pressure speed, low-pressure speed, and pressure ratio.

[0111] Furthermore, the expression for the pseudo-gradient time-varying parameter is:

[0112] , i=1,2,…,L y +L u ,

[0113] ,

[0114] in, Let be the pseudo-gradient time-varying parameters at time k. Let be the variable representing the pseudo-gradient time-varying parameter at time k. Let this be the first parameter variable at time k;

[0115] The expression for the equivalent data model is:

[0116] (3),

[0117] It is bounded at any time k.

[0118] Furthermore, obtaining the multi-step forward output prediction equation based on the equivalent data model includes:

[0119] make: ,

[0120] ,

[0121] ,

[0122] ,

[0123] , , , ,

[0124] in, Let this be the first intermediate variable at time k. Let k be the change in the system's output at time k. Let k be the change in the system output at time k-1. The second intermediate variable at time k. Let be the change in the system's input at time k. Let be the change in the system's input at time k-1. Let be the third intermediate variable at time k. Let A be the fourth intermediate variable at time k, and let B be the first matrix, C be the second matrix, and D be the fourth matrix.

[0125] Equation (3) can be rewritten as:

[0126] ,

[0127] The N-step forward prediction equation can be further given as follows:

[0128] ,

[0129] In the formula N u To control the time-domain constants and satisfy N u ≤N;

[0130] To fully reflect the dynamic characteristics of the controlled object, for time-delay systems such as engines, the prediction step size should be at least greater than the system's time-delay step size. Considering both online computational complexity and tracking performance, the prediction step size N=5 is set, and the control time-domain constant N is set. u =2;

[0131] make: ,

[0132] ,

[0133] ,

[0134] ,

[0135] ,

[0136] ,

[0137] ,

[0138] The expression for the multi-step forward output prediction equation is:

[0139] (4),

[0140] in, To predict output parameters, This is the fifth intermediate variable at time k. This is the fifth intermediate variable at time k+1. This is the sixth intermediate variable at time k. Let be the difference between the system input at time k+1 and the system input at time k. This is the first pseudo gradient matrix. This is the second pseudo-gradient matrix. This is the third pseudo-gradient matrix. This is the fourth pseudo-gradient matrix.

[0141] Furthermore, the online real-time estimation of the pseudo-gradient matrix in the multi-step forward output prediction equation based on the autoregressive and projection algorithms includes:

[0142] For the first pseudo-gradient matrix The second pseudo gradient matrix The third pseudo gradient matrix and the fourth pseudo gradient matrix The third intermediate variable at time k in and the fourth intermediate variable at time k The estimation is performed using a projection algorithm;

[0143] For the first pseudo-gradient matrix The second pseudo gradient matrix The third pseudo gradient matrix and the fourth pseudo gradient matrix The third intermediate variable at time k+i and the fourth intermediate variable at time k+i For i=1,2,3,4, a multi-level hierarchical prediction method is used for prediction.

[0144] In this embodiment, refer to Figure 2 The projected algorithm is used to estimate the pseudo gradient at the current time online, and the autoregressive model is used to estimate the pseudo gradient at the future time online to achieve dynamic linearization. The results are then input into the model-free adaptive predictive controller built in this application to predict the output parameters of the engine at future time and to solve the control parameters by rolling optimization. The optimal control parameters are obtained and input into the Adaptive Cycle Engine (ACE) for actual control. Based on the output of the ACE, dynamic linearization and optimization control are further performed, and finally, multi-variable precise cooperative control is achieved.

[0145] Furthermore, the expression for estimation using the projection algorithm is as follows:

[0146] (5),

[0147] in, Let μ be the change in the estimated value of the pseudo-gradient time-varying parameters, μ be the first step length factor, and η be the second step length factor, where μ > 0 and η ∈ (0, 1]. This represents the estimated pseudo-gradient time-varying parameter value at time k-1.

[0148] Furthermore, the expression for the auto-regressive (AR) model of the multi-level hierarchical prediction method is as follows:

[0149] (6),

[0150] ,

[0151] ,

[0152] (7),

[0153] in, The pseudo-gradient time-varying parameter estimate at time k is... For an unknown matrix, For the nth p There are n unknown matrix variables. p Let n be the order. p =2, This is an intermediate variable for estimating time-varying parameters of the pseudo-gradient;

[0154] Unknown matrix The formula for the least squares algorithm with a forgetting factor is determined by the following expression:

[0155] (8),

[0156] (9),

[0157] (10)

[0158] Where P(·) is the seventh intermediate variable, α(·) is the eighth intermediate variable, α0 is the second constant, and α0=0.99; when k=1, P(k-2)=P(-1)>0 and α(0)=0.95 are defined.

[0159] The following section describes the rolling optimization solution and optimal control command generation. A composite control strategy is constructed by embedding a model-free adaptive control algorithm based on single-step optimization into the rolling optimization framework of predictive control. Within this framework, a function (11) of the control input performance index with the goal of minimizing tracking error is designed, substituted into the multi-step forward output prediction equation, and then... By taking the derivative and setting its first derivative to zero, we can obtain equation (12).

[0160] Furthermore, the functional expression of the control input performance index is:

[0161] (11),

[0162] Where J is the control input performance index, and λ is a weighting factor used to limit the variation of the control input. This is the desired output signal.

[0163] Furthermore, obtaining the optimal control parameters includes:

[0164] Based on the control input performance index, the multi-step forward output prediction equation is substituted and the results are analyzed. Taking the derivative and setting its first derivative to zero, we obtain the following expression:

[0165] (12)

[0166] Using a rolling optimization strategy, only the first element of the sequence is applied to the system to obtain the optimal control parameters (the control input at the current moment), expressed as:

[0167] (13)

[0168] ,

[0169] Where I is the identity matrix and g is the fifth matrix.

[0170] For multi-input, multi-output nonlinear systems like adaptive cycle engines, the diagonal dominance condition of the system's input-output data relationship may be the only feasible option for describing the coupling between the system's variables. (Sub-blocks in PG) The diagonal dominance condition must be met, that is, the condition must be satisfied. , α≥1, ,and The signs of all elements in the expression remain unchanged for any k, j is the second variable, b1 is the first definite variable, and b2 is the second definite variable.

[0171] Therefore, an algorithm reset mechanism is introduced to enhance the PG estimation algorithm's ability to track time-varying parameters.

[0172] like or or i=1,2,3:

[0173] ;

[0174] like or i,j=1,2,3,i≠j:

[0175] ,

[0176] in, Let be the estimated value of the first parameter variable at time k, and sign(·) be the sign function.

[0177] In practice, based on the full-format dynamic linearized data model of the current operating point established online in step 1, the multivariable predictive control parameters designed in steps 2 and 3 are calculated autonomously in real time. A model-free adaptive predictive control (MFAPC) system suitable for the adaptive cycle engine is established, and multivariable control simulation verification is carried out for different modes (such as three-function (M2) and two-function (M3) modes) of a certain type of adaptive cycle engine at the ground operating point. The typical operating modes of this type of adaptive cycle engine are shown in Table 1.

[0178] Table 1 Typical operating modes of adaptive cycle engine

[0179]

[0180] In each simulation scenario, the sampling step size is set to 0.025s. At 1s, the high and low pressure speed and pressure ratio commands are simultaneously stepped by 4%. The initial value selection of the full-format dynamic linearized data model in the ground point dual-mode is shown in Equation (14). The initial value selection of the full-format dynamic linearized data model in the ground point triple-mode is shown in Equation (15).

[0181] (14)

[0182] (15).

[0183] Table 2 shows the main parameters of the MFAPC control algorithm under different operating modes of the ground point. In Table 2, diag(·) is a diagonal function. The curves showing the changes in the controller input and output parameters under the two-function and three-function modes are as follows: Figures 3a to 4b As shown in the figure, n H,r n represents the desired output value at high-pressure speed. L,r For the desired output value at low pressure and high speed, EPR r n is the expected output value of the pressure ratio. H (MFAPC) refers to the high-pressure speed based on the MFAPC control algorithm, n L (MFAPC) represents the low-pressure speed based on the MFAPC control algorithm, EPR(MFAPC) represents the pressure ratio based on the MFAPC control algorithm, and W fb (MFAPC) represents the main fuel flow rate based on the MFAPC control algorithm, and A8(MFAPC) represents the nozzle throat area based on the MFAPC control algorithm. RVABI (MFAPC) represents the area of ​​the adjustable duct ejector based on the MFAPC control algorithm.

[0184] Table 2 Main parameters of the MFAPC control algorithm

[0185]

[0186] Table 3 Ground Point Controller Step Response Performance Indicators

[0187]

[0188] As shown in Table 3, the algorithm exhibits excellent output tracking performance under different operating modes, with a step response settling time within 3.5 s, overshoot within 1%, and no steady-state error. This demonstrates that combining predictive control with model-free adaptive control can improve control performance and provides a solution for overcoming model uncertainty and achieving multivariable collaborative control.

[0189] Compared with existing model-free adaptive control algorithms, the model-free adaptive predictive control method (MFAPC) designed in this invention has a faster response speed and smaller overshoot, while fundamentally eliminating the dependence on precise mathematical models. By embedding the single-step optimization model-free adaptive mechanism into the rolling optimization framework of predictive control, the overall optimization capability of the system is significantly enhanced.

[0190] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A data-driven adaptive cyclic engine multivariable adaptive predictive control method, characterized in that, The method includes: Step 1: Based on the input and output data of the adaptive cycle engine, construct an equivalent data model of the adaptive cycle engine based on full-format dynamic linearization, and obtain the multi-step forward output prediction equation based on the equivalent data model. Step 2: Based on the autoregressive and projection algorithms, the pseudo gradient matrix in the multi-step forward output prediction equation is estimated online in real time to form a rolling optimization framework for predictive control. Step 3: Embed the model-free adaptive control algorithm based on single-step optimization into the rolling optimization framework to form a composite control strategy. Design the control input performance index with the goal of minimizing the tracking error and obtain the optimal control parameters. Step 4: Based on the composite control strategy, the optimal control parameters are calculated and generated online using the real-time input and output data of the adaptive cycle engine, so as to achieve precise and coordinated multi-variable control of the engine. The step of obtaining the multi-step forward output prediction equation based on the equivalent data model includes: make: , ， ， ， ， ， ， ， Where R is the set of real numbers, Let this be the first intermediate variable at time k. Let k be the change in the system's output at time k. Let k be the change in the system output at time k-1. The second intermediate variable at time k. Let k be the change in the system's input at time k. Let be the change in system input at time k-1. Let be the third intermediate variable at time k. Let A be the fourth intermediate variable at time k, and let B be the first matrix, C be the second matrix, and D be the fourth matrix. Set the prediction step size N=5, and set the control time domain constant N. u =2; make: , ， ， ， ， ， ， The expression for the multi-step forward output prediction equation is: ， in, To predict output parameters, This is the fifth intermediate variable at time k. This is the fifth intermediate variable at time k+1. This is the sixth intermediate variable at time k. Let be the difference between the system input at time k+1 and the system input at time k. This is the first pseudo gradient matrix. This is the second pseudo-gradient matrix. This is the third pseudo-gradient matrix. This is the fourth pseudo-gradient matrix; The online real-time estimation of the pseudo-gradient matrix in the multi-step forward output prediction equation based on autoregression and projection algorithms includes: For the first pseudo-gradient matrix The second pseudo gradient matrix The third pseudo gradient matrix and the fourth pseudo gradient matrix The third intermediate variable at time k in and the fourth intermediate variable at time k The estimation is performed using a projection algorithm; For the first pseudo-gradient matrix The second pseudo gradient matrix The third pseudo gradient matrix and the fourth pseudo gradient matrix The third intermediate variable at time k+i and the fourth intermediate variable at time k+i For i=1,2,3,4, a multi-level hierarchical prediction method is used for prediction.

2. The data-driven adaptive cyclic engine multivariable adaptive predictive control method according to claim 1, characterized in that, The construction of the equivalent data model of the adaptive cyclic engine based on full-format dynamic linearization includes: Step 11: Based on the analysis of the mapping relationship between the opening degree of the variable geometry components of the adaptive cycle engine and the overall engine performance, determine the control variables and controlled parameters, and construct a multivariable control framework reflecting the internal energy conversion and thrust output of the engine. The expression is: ， Where u(k) is the input of the system at time k, y(k) is the output of the system at time k, and n u Input an unknown order, n, into the system. y The system output has an unknown order, and f(·) is a nonlinear function; Step 12, Definition To input the relevant sliding time window All control input signals within and in the output-related sliding time window The first vector is composed of all system output signals within it. The expression is: ， ， When k≤0 is satisfied , Where m is the dimension. Let be the first pseudo-order of the system. The second pseudo-order of the system, ; Step 13: When the expression of the multivariate control framework satisfies conditions one and two, then when At that time, there exists a pseudo-gradient time-varying parameter that transforms the first vector into an equivalent data model. The first condition is: the nonlinear function f of the object i (·) has continuous partial derivatives with respect to the components of each variable, where i is the first variable, i=1,2,…,m; The second condition is: for any two times k1 and k2... , ,have ω is the first constant greater than 0. This represents the change in the first vector at time k. This represents the change in the system's output at time k+1.

3. The data-driven adaptive cyclic engine multivariable adaptive predictive control method according to claim 2, characterized in that, The controlled quantities include the main fuel flow rate, the tailpipe throat area, and the area of ​​the rear adjustable duct ejector; the controlled parameters include the high-pressure speed, the low-pressure speed, and the pressure ratio.

4. The data-driven adaptive cyclic engine multivariable adaptive predictive control method according to claim 2, characterized in that, The expression for the pseudo-gradient time-varying parameter is: , ,i=1,2,…,L y +L u , ， in, Let be the pseudo-gradient time-varying parameters at time k. Let be the variable representing the pseudo-gradient time-varying parameter at time k. Let this be the first parameter variable at time k; The expression for the equivalent data model is: ， It is bounded at any time k.

5. The data-driven adaptive cyclic engine multivariable adaptive predictive control method according to claim 4, characterized in that, The expression for estimation using the projection algorithm is as follows: ， in, Let μ be the change in the estimated pseudo-gradient time-varying parameters at time k, μ be the first step length factor, and η be the second step length factor, where μ > 0 and η ∈ (0, 1]. This represents the estimated pseudo-gradient time-varying parameter value at time k-1.

6. The data-driven adaptive cyclic engine multivariable adaptive predictive control method according to claim 5, characterized in that, The expression for the autoregressive model of the multi-level hierarchical prediction method is as follows: ， ， ， ， in, The pseudo-gradient time-varying parameter estimate at time k is... For an unknown matrix, For the nth p There are n unknown matrix variables. p For order, This is an intermediate variable for estimating time-varying parameters of the pseudo-gradient; Unknown matrix The formula for the least squares algorithm with a forgetting factor is determined by the following expression: ， ， ， Where P(·) is the seventh intermediate variable, α(·) is the eighth intermediate variable, α0 is the second constant, and α0=0.99; when k=1, P(k-2)=P(-1)>0 and α(0)=0.95 are defined.

7. The data-driven adaptive cyclic engine multivariable adaptive predictive control method according to claim 6, characterized in that, The functional expression of the control input performance index is: ， Where J is the control input performance index, and λ is the weighting factor. This is the desired output signal.

8. The data-driven adaptive cyclic engine multivariable adaptive predictive control method according to claim 7, characterized in that, Obtaining the optimal control parameters includes: Based on the control input performance index, the multi-step forward output prediction equation is substituted and the results are analyzed. Taking the derivative and setting its first derivative to zero, we obtain the following expression: ， The optimal control parameters are obtained by employing a rolling optimization strategy, expressed as follows: ， ， Where I is the identity matrix and g is the fifth matrix.

Images