Aero-engine multivariable robust adaptive control method and system

Abstract

The application discloses an aero-engine multivariable robust adaptive control method. The method combines an augmented linear quadratic regulation control method with a model reference adaptive control method, dynamically and adaptively compensates the output of the augmented linear quadratic regulation control loop by the output of the model reference adaptive control loop, and judges whether to start the adaptive compensation according to a control error. The application further discloses an aero-engine multivariable robust adaptive control system. Compared with the prior art, the application can improve the dynamic response characteristic of the control system at a non-design point without changing the steady-state control performance.

Description

Technical Field

[0001] This invention relates to a multivariable robust adaptive control method and system for aero-engines, belonging to the field of aero-engine control technology. Background Technology

[0002] Aero engines possess highly complex structures and closely integrated components, exhibiting a wide range of parameter variations within their operating envelope and operating in highly variable environments, resulting in strong nonlinearity and uncertainty. Furthermore, engines operate under prolonged conditions of high temperature, high pressure, and strong vibration. Due to wear, fouling, and other factors, the thermal efficiency and flow capacity of rotating components can change, causing engine characteristics to deviate from the design parameters and introducing additional uncertainties. Therefore, engine control systems must possess robustness and adaptive capabilities.

[0003] Meanwhile, engines have evolved into multivariable control systems. Especially with the development of variable cycle engines, combined cycle engines, and short takeoff and vertical landing engines, single-variable control methods are no longer sufficient to meet control requirements, necessitating the adoption of multivariable control methods to improve control performance. The multivariable linear quadratic regulator (LQR) is simple in structure, possesses infinite amplitude margin and a phase margin greater than 60°, and is a classic multivariable control method. The augmented LQR control method (ALQR), by adding an integrator, gives the system good steady-state robustness. However, when applied to engine control, due to its uncertainty and large parameter variation range, it cannot achieve good dynamic performance throughout the entire envelope and life cycle.

[0004] Adaptive control is a nonlinear control method primarily used to handle uncertain systems. When faced with unmodeled dynamics caused by factors such as engine degradation, the dynamic performance of multivariable robust controllers designed based on nominal models deteriorates. Adaptive control, however, can reduce the impact of system uncertainties on the dynamic performance of the control system by updating controller parameters in real time. Therefore, combining adaptive control with augmented LQR control can improve engine dynamic performance while satisfying steady-state performance requirements. Current research on multivariable adaptive control methods faces challenges such as numerous adaptive adjustable parameters or complex control structures, which introduce new uncertainties into the adaptive adjustment of loop gain. Summary of the Invention

[0005] The technical problem to be solved by this invention is to overcome the shortcomings of the prior art and provide a multivariable robust adaptive control method for aero-engines. This method organically combines the ALQR control method with the model reference adaptive control method and performs dynamic adaptive compensation on the ALQR control results based on the control error. This can improve the dynamic response characteristics of the controller at non-design points without changing the steady-state control performance.

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

[0007] A robust multivariable adaptive control method for aero-engines uses a model reference adaptive control loop to dynamically compensate the main control output of the main control loop to generate the aero-engine control input:

[0008] In the main control loop, the output tracking error e between the control command r and the engine output y is... y Integrate, e y After integration, the engine state variable x, which includes the actuator, is compared with the state variable x. p Formation of error augmented state quantity x a x a The main control output u is obtained through the main controller. main The main controller is based on x a As an augmented linear quadratic control controller for state variables;

[0009] In the model reference adaptive control loop, the control command r is input to the reference state quantity x generated by the reference model. m With the aforementioned error augmentation state quantity x a The tracking error e of the reference model state variable is obtained by subtraction, and e is judged. If e is greater than or equal to a preset error threshold, the error is augmented to the state variable x. a The reference model's state tracking error e and regression vector Φ(x) are input to the adaptive controller and output by the adaptive controller u. ad For the main control output u main Compensation will be performed; otherwise, the main control output u will not be compensated. main Compensation is performed; the reference model of the model reference adaptive control loop is a closed-loop linear nominal control system based on the augmented linear quadratic regulating controller, and the adaptive controller of the model reference adaptive control loop is an e-corrected robust adaptive controller.

[0010] Based on the same inventive concept, the following technical solutions can also be obtained:

[0011] A multivariable robust adaptive control system for an aero-engine includes a main control loop and a model reference adaptive control loop. The model reference adaptive control loop is used to perform the following dynamic compensation on the main control output of the main control loop to generate the aero-engine control input:

[0012] In the main control loop, the output tracking error e between the control command r and the engine output y is... y Integrate, e y After integration, the engine state variable x, which includes the actuator, is compared with the state variable x. p Formation of error augmented state quantity x a xa The main control output u is obtained through the main controller. main The main controller is based on x a As an augmented linear quadratic control controller for state variables;

[0013] In the model reference adaptive control loop, the control command r is input to the reference state quantity x generated by the reference model. m With the aforementioned error augmentation state quantity x a The tracking error e of the reference model state variable is obtained by subtraction, and e is judged. If e is greater than or equal to a preset error threshold, the error is augmented to the state variable x. a The reference model's state tracking error e and regression vector Φ(x) are input to the adaptive controller and output by the adaptive controller u. ad For the main control output u main Compensation will be performed; otherwise, the main control output u will not be compensated. main Compensation is performed; the reference model of the model reference adaptive control loop is a closed-loop linear nominal control system based on the augmented linear quadratic regulating controller, and the adaptive controller of the model reference adaptive control loop is an e-corrected robust adaptive controller.

[0014] Preferably, the controller gain of the augmented linear quadratic regulating controller is optimized using a metaheuristic optimization method based on Newton-Raphson, with the goal of minimizing the tracking error.

[0015] Furthermore, during the dynamic compensation process, when switching from compensation to no compensation, the following method is used for perturbation-free switching: the integral of the output tracking error at time k is calculated as follows: yI Set to zero, and use u(k) = u0 + u main (k) represents the control variable for the aero-engine at time k, where u0 = u(k-1) = u main (k-1)+u ad (k-1) represents the control input of the aero-engine at the previous moment, u main (k) represents the main control output at time k.

[0016] Compared with the prior art, the technical solution of the present invention and its further preferred and improved solutions have the following beneficial effects:

[0017] The adaptive control strategy is adopted only during dynamic processes, resulting in high stability. Conventional adaptive compensation control methods adapt to changes in engine characteristics by adjusting the parameters of the main control loop controller. However, changes in gain directly affect the stability of the system and introduce new uncertainties. This invention compensates only for the output of the main controller during dynamic processes without changing the parameters of the main controller, thus maintaining the robustness of the original augmented LQR control.

[0018] The reference model exhibits good stability and dynamic performance, providing an appropriate tracking target for adaptive control. This invention employs a novel Newton-Raphson metaheuristic algorithm to optimize the nominal model controller parameters. The closed-loop reference model obtained based on the optimized parameters possesses the expected dynamic and steady-state performance, providing a good reference model for adaptive control.

[0019] The adaptive control transitions smoothly during entry and exit. This invention employs a bumpless switching control method, which re-initializes the integrator and the initial control input of the controller when switching to the main controller. Even with a large adaptive compensation control input, this ensures a smooth switching process. Attached Figure Description

[0020] Figure 1 A schematic diagram of a twin-shaft turbofan engine.

[0021] Figure 2 Performance index transformation curves when optimizing controller parameters for the Newton-Raphson metaheuristic algorithm;

[0022] Figure 3 To augment the closed-loop step response curve of the nominal model of the LQR controller. Where (a) represents N... l Step response curve; (b) is π epr Step response curve; (c) is π cepr Step response curve;

[0023] Figure 4 This is a diagram of the control system structure.

[0024] Figure 5 The controlled variable response curve is simulated for a linear system, where (a) is N. l Step response curve; (b) is π epr Step response curve; (c) is π cepr Step response curve;

[0025] Figure 6 The simulation curves of the control quantity change for a linear system are shown, where (a) represents W. fm (a) is the curve of change; (b) is the curve of change of A8; (c) is the curve of change of A. rvabi Change curve;

[0026] Figure 7 The simulation of the controlled variable response curve for a nonlinear system (H = 0 km, Ma = 0), where (a) is N l Step response curve; (b) is π epr Step response curve; (c) is π cepr Step response curve;

[0027] Figure 8The simulation curve of the control quantity change for the nonlinear system (H = 0km, Ma = 0);

[0028] Figure 9 The nonlinear simulation controlled variable response curve (H = 1.2 km, Ma = 0.8) is shown, where (a) is N. l Step response curve; (b) is π epr Step response curve; (c) is π cepr Step response curve;

[0029] Figure 10 The curve showing the change in control input during simulation of a nonlinear system (H = 1.2 km, Ma = 0.8). Detailed Implementation

[0030] To address the shortcomings of existing technologies, the present invention organically combines the ALQR control method with the model reference adaptive control method. The model reference adaptive control loop output is used to dynamically and adaptively compensate the ALQR control loop output, and the adaptive compensation is activated based on the control error. This improves the dynamic response characteristics of the controller at non-design points without changing the steady-state control performance.

[0031] The specific technical solution proposed in this invention is as follows:

[0032] A robust multivariable adaptive control method for aero-engines uses a model reference adaptive control loop to dynamically compensate the main control output of the main control loop to generate the aero-engine control input:

[0033] In the main control loop, the output tracking error e between the control command r and the engine output y is... y Integrate, e y After integration, the engine state variable x, which includes the actuator, is compared with the state variable x. p Formation of error augmented state quantity x a x a The main control output u is obtained through the main controller. main The main controller is based on x a As an augmented linear quadratic control controller for state variables;

[0034] In the model reference adaptive control loop, the control command r is input to the reference state quantity x generated by the reference model. m With the aforementioned error augmentation state quantity x a The tracking error e of the reference model state variable is obtained by subtraction, and e is judged. If e is greater than or equal to a preset error threshold, the error is augmented to the state variable x. a The reference model's state tracking error e and regression vector Φ(x) are input to the adaptive controller and output by the adaptive controller u. adFor the main control output u main Compensation will be performed; otherwise, the main control output u will not be compensated. main Compensation is performed; the reference model of the model reference adaptive control loop is a closed-loop linear nominal control system based on the augmented linear quadratic regulating controller, and the adaptive controller of the model reference adaptive control loop is an e-corrected robust adaptive controller.

[0035] A multivariable robust adaptive control system for an aero-engine includes a main control loop and a model reference adaptive control loop. The model reference adaptive control loop is used to perform the following dynamic compensation on the main control output of the main control loop to generate the aero-engine control input:

[0036] In the main control loop, the output tracking error e between the control command r and the engine output y is... y Integrate, e y After integration, the engine state variable x, which includes the actuator, is compared with the state variable x. p Formation of error augmented state quantity x a x a The main control output u is obtained through the main controller. main The main controller is based on x a As an augmented linear quadratic control controller for state variables;

[0037] In the model reference adaptive control loop, the control command r is input to the reference state quantity x generated by the reference model. m With the aforementioned error augmentation state quantity x a The tracking error e of the reference model state variable is obtained by subtraction, and e is judged. If e is greater than or equal to a preset error threshold, the error is augmented to the state variable x. a The reference model's state tracking error e and regression vector Φ(x) are input to the adaptive controller and output by the adaptive controller u. ad For the main control output u main Compensation will be performed; otherwise, the main control output u will not be compensated. main Compensation is performed; the reference model of the model reference adaptive control loop is a closed-loop linear nominal control system based on the augmented linear quadratic regulating controller, and the adaptive controller of the model reference adaptive control loop is an e-corrected robust adaptive controller.

[0038] To facilitate public understanding, the technical solution of this invention will be described in detail below using the cruise engine of the STOVL propulsion system, namely a twin-shaft turbofan engine, as an example, in conjunction with the accompanying drawings:

[0039] The structure of a twin-shaft turbofan engine is as follows Figure 1As shown, it includes a fan, bypass, compressor, combustion chamber, high-pressure turbine, low-pressure turbine, mixing chamber, afterburner, and exhaust nozzle. Compared to conventional turbofan engines, it features added adjustable geometry, including a rear variable area bypass injector (RVABI) and a low-pressure turbine inlet guide vane.

[0040] The output variable of the selected control system is the low-pressure speed N. l Engine pressure ratio π epr Engine core pressure ratio π cepr The input variable is the fuel flow rate W. fm Tail nozzle throat area A8, RVABI opening A rvabi The state variable is the low-pressure speed N. l High pressure and high speed N h High-pressure turbine after-temperature T 42 .

[0041] Considering the existence of input uncertainty and matching uncertainty in the system, then we have

[0042]

[0043] Where, x = [N l N h T 42 W fm A8 A rvabi ] T To augment the state variables after the actuator, the input variable is u = [W fm,r A 8,r A rvabi,r ] T The subscript 'r' represents the instruction signal for the corresponding variable, and the output quantity is y = [N]. l π epr π cepr ] T Λ is an unknown diagonal matrix, where the diagonal elements represent the uncertainty in the effectiveness of the control input. Θ *T Φ(x) represents the parameter matching uncertainty, and Θ * Let be an unknown constant matrix, and Φ(x) be a known bounded regression function.

[0044] To ensure robust tracking performance of the controller, the output tracking error is augmented into the state variables, facilitating zero steady-state error tracking. The output tracking error can be expressed as...

[0045]

[0046] Where r represents the control command.

[0047] By augmenting the integral of the output error in equation (2) into the state variables in equation (1), the augmented state-space model can be obtained.

[0048]

[0049] in, C a =[C 0], D a =D.

[0050] This invention employs the augmented LQR method to eliminate steady-state error. The control command is idealized as a step command, whose derivative is 0. Therefore, taking the nominal model of equation (3) and differentiating it, we have:

[0051]

[0052] in,

[0053] The state feedback controller is designed based on equation (4), and the performance functional is expressed as follows:

[0054]

[0055] Among them, Q c and R c For a weighted matrix of appropriate dimension, and Q c =Q c T ≥0, R c =R c T >0.

[0056] The control law u is obtained based on the LQR control method. c =K c x c ,and Where P c Obtained from the Riccati equation

[0057]

[0058] Then there is a control law

[0059]

[0060] Wherein, the controller gain is

[0061] The controller gain of an ALQR controller typically needs to be optimized using an optimization algorithm. Currently, widely used optimization algorithms are mainly divided into gradient-based and population-based algorithms. Gradient-based optimization algorithms are prone to getting trapped in local optima, while population-based optimization algorithms are better at finding the global optimum. However, population-based optimization algorithms suffer from slow convergence and difficulty in parameter tuning. Therefore, this embodiment uses a Newton-Raphson metaheuristic optimization (NRBO) algorithm to optimize the controller gain, combining gradient-based and population-based optimization algorithms to overcome the shortcomings of both.

[0062] The NRBO algorithm is mainly based on two principles: the Newton-Raphson search rule and the trap avoidance operator, to search and avoid getting trapped in local optima. The main optimization steps are as follows:

[0063] Step A. Population initialization.

[0064] Generate an initial random population within the boundary of the candidate solutions:

[0065]

[0066] Where n = 1, 2, ..., N p j = 1, 2, ..., dim, x j n Let represent the position of the nth individual in the j-th dimension, lb represent the lower bound of the candidate solutions, ub represent the upper bound of the candidate solutions, and rand represent a random number between (0,1). Therefore, the population matrix can be described as...

[0067]

[0068] Step B. Newton-Raphson search rule.

[0069] The search rule is a crucial part of NRBO, primarily based on Newton's iteration method, which is an iterative approximation solution. Furthermore, NRBO uses the current position variable instead of the fitness value to update variables, reducing computation time. The update rule for the nth individual in generation t+1 is as follows:

[0070]

[0071] in

[0072]

[0073] in, X represents the Hadamard product, where randn is a normally distributed random number with a mean of 0 and a variance of 1. w X represents the worst individual at present. b Represents the current optimal individual, Mean() represents the mean, and n1 and n2 represent the values ​​in (1, N).p The integers are random integers between (0, 1) and (n1, 1), and n1 ≠ n2. a and b are random numbers between (0, 1) and t. max This represents the maximum number of iterations, and r1 and r2 represent random numbers between (0,1).

[0074] Step C. Trap avoidance operator.

[0075] Traps avoidance operators can help NRBO avoid getting trapped in local optima, improving its effectiveness in handling practical problems. Through trap avoidance operators, the optimal individual X can be combined with a certain probability. b This results in a higher quality individual.

[0076]

[0077] Where DF is the probability of executing the trap avoidance operator, θ1 and θ2 are random numbers between (-1,1) and (-0.5,0.5) respectively, and the random numbers μ1 = 3β×rand + (1-β) and μ2 = β×rand + (1-β).

[0078] Therefore, to achieve good dynamic tracking performance, this embodiment takes minimizing the tracking error as the optimization objective and uses NRBO to optimize Q. c R c Matrix parameters. The optimization objective can be expressed as...

[0079]

[0080] Where R is the time series of the input signal and Y is the time series of the output signal.

[0081] Input signal N 1,r π epr,r π cepr,r Set the unit step response output of the inertial system with a time constant of 0.8, and set Q... c R c Simplifying the matrix to a diagonal matrix, there are 12 parameters to be optimized, corresponding to Q in sequence. c R c The diagonal elements of the matrix. Before optimization, the parameters were set with an upper bound of 100 and a lower bound of 10. -4 After 300 generations of optimization, the performance metrics change curves are as follows: Figure 2 As shown.

[0082] Depend on Figure 2 It can be observed that as the number of iterations increases, the performance index gradually decreases, the tracking error gradually diminishes, and the controller performance is continuously optimized. Based on the optimal Q... c R c Matrix parameters are used for controller design to obtain the ALQR controller gain K.c Based on the nominal model, the unit step response of the ALQR controller closed-loop system was calculated, and the response curve is shown below. Figure 3 As shown. Figure 3 As shown, the system optimized with NRBO has a rapid response, good tracking of the reference signal, and no significant errors in steady state or dynamics.

[0083] ALQR is designed for nominal models. When engine parameters change due to factors such as operating point variations or performance degradation, the controller parameters no longer match the target, leading to a deterioration in the system's dynamic characteristics. This invention uses Model Reference Adaptive Control (MRAC) to compensate for these parameter changes. Considering that standard MRAC control may generate high gain when facing sufficiently large disturbances, making it impossible to determine the boundaries of the adaptive parameters and ultimately causing system instability, this invention employs an e-correction method to add a damping term to constrain the adaptive parameters. The damping term is reduced proportionally to the norm of the tracking error; when the tracking error approaches zero, the damping term also decreases to zero.

[0084] The nominal model is obtained from equation (3) as follows:

[0085]

[0086] At the same time, from equation (7) we get Let u=u main Then refer to the model

[0087] It can be represented as

[0088]

[0089] Among them, A m =A a +B a K c B m =B r C m =C a +D a K c .

[0090] The adaptive compensation control quantity u ad Introducing equation (3), we have

[0091]

[0092] make An adaptive controller can then be designed.

[0093] u ad =-K ad x a -Θ TΦ(x) (17)

[0094] Among them, K ad And Θ are opposites and Θ * If the estimated value is , then let the estimation error be .

[0095] Substituting equation (17) into equation (16), we get

[0096]

[0097] The closed-loop tracking error equation is obtained as follows:

[0098]

[0099] To enhance the robustness of the adaptive control system, the robust adaptive law modified by e is taken as follows:

[0100]

[0101] in, P is the adaptive rate matrix. e =P e T >0 represents the Lyapunov equation. The unique solution, where Q e =Q e T >0. η1>0 and η2>0 are correction coefficients.

[0102] Choose Lyapunov functions as

[0103]

[0104] Differentiating both sides, we get

[0105]

[0106] Then there is

[0107]

[0108] achievable

[0109]

[0110] Where λ min () represents the smallest eigenvalue of the matrix.

[0111] To ensure Then there is

[0112]

[0113] or

[0114]

[0115] Based on the above analysis, the robust adaptive control method based on e-correction ensures the consistent eventual boundedness of the error e and also guarantees the boundedness of the adaptive gain. When the conditions described in equation (25) or (26) are met, the robust adaptive control method based on e-correction can improve the dynamic performance of the control system while satisfying stability. When the lower bound of equation (25) or (26) is reached, the adaptive compensation control is turned off, and only ALQR control is relied upon. The switching logic is set according to the state tracking error e.

[0116]

[0117] To achieve a seamless handover when the adaptive loop exits operation, this invention further employs the following method for seamless handover when switching from compensation to no compensation during dynamic compensation: The integral of the output tracking error e at the current time k is... yI Set to zero, and use u(k) = u0 + u main (k) represents the control variable for the aero-engine at time k, where u0 = u(k-1) = u main (k-1)+u ad (k-1) represents the control input of the aero-engine at the previous moment, u main (k) represents the main control output at time k. The switching logic for this bumpless handover can be expressed as:

[0118]

[0119] Ultimately formed as Figure 4 The multivariable adaptive control system for aero-engines shown in this invention includes a main control loop and a model reference adaptive control loop. For example... Figure 4 As shown, the input control command r enters the control system; in the main control loop, the output tracking error e is obtained from the command r and the engine output y. y e y The state variable x after integration and augmentation with the engine actuator p Formation of error augmented state quantity x a x a The main control output u is obtained through the main controller ALQR. main =K c x a With adaptive compensation output u ad Obtain engine control input u = u main +u ad, u then acts on the engine controlled object; in the model reference adaptive control loop, the command r is input into the reference model, which is designed as a closed-loop linear nominal control system optimized by the ALQR controller. The reference state quantity x output by the reference model m is subtracted from the actual state quantity x a to obtain the reference model state quantity tracking error e. Then, e is judged. If ∥e∥≥p, the actual state quantity x a , the reference model state quantity tracking error e, and the regression vector Φ(x) are substituted into the adaptive law to obtain the adaptive parameters K ad and Θ, and then the adaptive compensation output u ad =-K ad x a -Θ T Φ(x); if ∥e∥<p, then let u ad =0, that is, no adaptive compensation is performed.

[0120] In the Figure 4 structure, the bounded tracking caused by the correction of e has little impact on the practical application of the control system. Since the error is easy to calculate, it is judged by Equation (25). When ∥e∥≥p, the ALQR controller is dynamically compensated through robust adaptive control to improve the dynamic characteristics of the ALQR controller. Otherwise, only rely on ALQR for steady-state control to maintain the good steady-state performance of ALQR.

[0121] The ALQR control, model reference adaptive control, robust adaptive control method for e correction, etc. involved in the technical solution of the present invention are all existing technologies, and those skilled in the art can design and implement them according to the targeted aeroengine.

[0122] In order to verify the effect of the technical solution of the present invention, an engine control system as shown in Figure 4 is built to carry out simulation verification.

[0123] First, carry out simulation verification based on the linear model. Set Q e as a 9th-order identity matrix, the adaptation rate Γ K =diag([1,145,1,1,500,1,1,485,1]), Γ Θ =diag([1,1,110]), and the correction parameters η1 = η2 = 0.3. Take Θ * Φ(x)=m·[1,1,1;1,1,1;1,1,1]·[N l N h T 42 TWhere m represents the range of uncertainty, and m = 0 represents the nominal model. To verify the adaptive control effect, unit step simulations of the linear system were performed with m = 0.5 and m = 0.8 respectively. The controlled variable response curves of the linear system are shown below. Figure 5 As shown in the figure, "ALQR" represents the simulation result of the ALQR control system without adaptive strategy, and "ALQR+RAC" represents the simulation result of the ALQR control system based on robust adaptive dynamic compensation. The specific dynamic performance indicators of the linear system simulation when m=0.5 and m=0.8 are shown in Table 1 and Table 2.

[0124] Table 1. Dynamic performance indicators of linear simulation when m = 0.5

[0125]

[0126] Table 2 shows the dynamic performance indicators of linear simulation when m = 0.8.

[0127]

[0128] From Table 1, Table 2 and Figure 5 It is evident that when unmodeled dynamics exist, the controlled variable response of the ALQR control system exhibits overshoot and changes in response speed. In the simulation using the nominal model, the overshoot of all three controlled variables is zero, and the settling time is t. s =2.40s. As the unmodeled dynamics increase, significant differences emerge in the responses of each loop in a simple ALQR control system. The unmodeled dynamics cause the low-pressure speed N to... l and core engine pressure ratio π cepr The response speed is faster, and the engine pressure ratio π epr The response speed slows down, while the response speed speed increases, resulting in overshoot, especially when m = 0.8 and π. epr The overshoot reached 5.93%, and the speed overshoot also reached 2.49%, failing to meet the control targets. epr The response speed is significantly slower, with the adjustment time exceeding 6 seconds in both cases, which does not meet the control requirements.

[0129] However, ALQR+RAC control based on robust adaptive dynamic compensation, when faced with unmodeled dynamics, N l and π cepr The response settling time is between 2.32s and 2.44s, which is not significantly different from the nominal system. eprThe response time is faster, between 1.84 and 2.26 s. In both unmodeled dynamic scenarios, the ALQR+RAC control method showed no overshoot, verifying that the robust adaptive compensation control effectively compensated for the impact of unmodeled dynamics on the system's dynamic performance. When ∥e∥ reached the lower bound of equation (25), adaptive compensation was disabled, and steady-state control was performed solely through ALQR. This also solved the bounded tracking problem inherent in the robust adaptive controller based on e-correction, maintaining the good steady-state performance of ALQR. It is worth noting that a disturbance-free switching control strategy was adopted between the dynamic ALQR+RAC and the steady-state ALQR control, ensuring a smooth transition between the control and controlled variables of the two controllers.

[0130] The curve of the change of control quantity in linear simulation is as follows: Figure 6 As shown in the figure, "ALQR" represents the ALQR control system, "ALQR+RAC(u)" represents the total output u of ALQR+RAC, and "RAC(u)" represents the total output u of ALQR+RAC. ad ")" indicates the adaptive compensation output u in ALQR+RAC. ad .

[0131] Depend on Figure 6 It was found that in the simulations of the ALQR control system based on robust adaptive dynamic compensation at m=0.5 and m=0.8, the system shuts off the robust adaptive compensation control (u) at 7.28s and 6.74s, respectively, because ∥e∥ is sufficiently small and close to steady state. ad =0), relying solely on ALQR for control.

[0132] The dynamic change process of the control quantity in the ALQR control system is different from that in the ALQR control system based on robust adaptive dynamic compensation, but the final steady-state value is the same, and both achieve good steady-state performance.

[0133] Based on the component-level model of the F135-PW-600 engine, nonlinear simulations were performed to verify both the ALQR controller and the ALQR+RAC controller based on robust adaptive dynamic compensation. According to the flight envelope of the F-35B aircraft, simulations were conducted at operating points with flight conditions of H=0km, Ma=0 and H=1.2km, Ma=0.8, respectively. A 1% step change was applied to the controlled variable command, and the response curves are shown below. Figures 7 to 10 As shown, the specific performance indicators are shown in Tables 3 and 4.

[0134] Table 3 Dynamic performance indicators when H=0km, Ma=0

[0135]

[0136] Depend on Figure 7As can be seen from Table 3, at the design point, the control performance of ALQR and ALQR+RAC is similar, except that π under ALQR control... cepr A slight overshoot occurs; the speed N under ALQR+RAC control... l π epr The response time is fast, all below 2.1 seconds, π cepr The response is slower, but still slightly faster than π under ALQR control. epr The response time of the three loops differs by 0.86s under ALQR control. After adopting adaptive control, the difference in the settling time of the three loops is reduced to 0.68s, a reduction of 20.93%, which is beneficial to the stability of the system and avoids oscillations or instability caused by the fast loop compensating for changes in the slow loop in practical applications. Furthermore, the dynamic response process of ALQR+RAC is more stable; under ALQR control, the π-time difference is within 1 second. epr There is a fluctuation of approximately 12% in amplitude, π cepr There is a fluctuation of 1.68%. The controlled variable responds smoothly in the ALQR+RAC control system, maintaining the good steady-state performance of ALQR and achieving zero steady-state error tracking. Furthermore, due to... Figure 8 It can be observed that after adaptive dynamic compensation, the control quantity W fm The fluctuations decreased, achieving relatively stable changes.

[0137] The controlled variable response curve at the operating point H = 1.2 km and Ma = 0.8 is as follows: Figure 9 As shown, the control quantity change curve is as follows: Figure 10 As shown in Table 4, the dynamic performance indicators are as follows.

[0138] Table 4 Dynamic performance indicators at H = 1.2 km and Ma = 0.8

[0139]

[0140] Depend on Figure 9 As shown in Table 4, at non-design points, due to the existence of unmodeled dynamic responses, the dynamic response performance of the controlled variable in the ALQR control system deteriorates, and π... epr π cepr Overshoot reached 3.75% and 2.50%. N l The response time slowed down to 3.48 seconds, π. epr π cepr The response speed is faster, with the adjustment time of the three loops differing by 1.84 seconds. After adopting adaptive control, π... cepr The response speed is slower, with the settling times of the three loops differing by 0.88 seconds, a 52.17% reduction compared to ALQR. Meanwhile, due to... Figure 10 It can be observed that the changes in the control quantity become more stable after adaptive compensation, which is beneficial to the stability of the system.

Claims

1. A multivariable robust adaptive control method for aero-engines, characterized in that, The main control output of the main control loop is dynamically compensated using a model reference adaptive control loop to generate the control input for the aero-engine: In the main control loop, the output tracking error e between the control command r and the engine output y is... y Integrate, e y After integration, the engine state variable x, which includes the actuator, is compared with the state variable x. p Formation of error augmented state quantity x a x a The main control output u is obtained through the main controller. main The main controller is based on x a As an augmented linear quadratic control controller for state variables; In the model reference adaptive control loop, the control command r is input to the reference state quantity x generated by the reference model. m With the error augmentation state quantity x a The tracking error e of the reference model state variable is obtained by subtraction, and e is judged. If e is greater than or equal to a preset error threshold, the error is augmented to the state variable x. a The reference model's state tracking error e and regression vector Φ(x) are input to the adaptive controller and output by the adaptive controller u. ad For the main control output u main Compensation will be performed; otherwise, the main control output u will not be compensated. main Compensation is performed; the reference model of the model reference adaptive control loop is a closed-loop linear nominal control system based on the augmented linear quadratic regulating controller, and the adaptive controller of the model reference adaptive control loop is an e-corrected robust adaptive controller.

2. The multivariable robust adaptive control method for aero-engines as described in claim 1, characterized in that, The controller gain of the augmented linear quadratic regulator is optimized using a metaheuristic optimization method based on Newton-Raphson, with the goal of minimizing the tracking error.

3. The multivariable robust adaptive control method for aero-engines as described in claim 1, characterized in that, When switching from compensation to no compensation during dynamic compensation, the following method is used for perturbation-free switching: the integral of the output tracking error at time k is calculated as follows: yI Set to zero, and use u(k) = u0 + u main (k) represents the control variable for the aero-engine at time k, where u0 = u(k-1) = u main (k-1)+u ad (k-1) represents the control input of the aero-engine at the previous moment, u main (k) represents the main control output at time k.

4. A multivariable robust adaptive control system for an aero-engine, characterized in that, It includes a main control loop and a model reference adaptive control loop, wherein the model reference adaptive control loop is used to perform the following dynamic compensation on the main control output of the main control loop to generate the control input for the aero-engine: In the main control loop, the output tracking error e between the control command r and the engine output y is... y Integrate, e y After integration, the engine state variable x, which includes the actuator, is compared with the state variable x. p Formation of error augmented state quantity x a x a The main control output u is obtained through the main controller. main The main controller is based on x a As an augmented linear quadratic control controller for state variables; In the model reference adaptive control loop, the control command r is input to the reference state quantity x generated by the reference model. m With the error augmentation state quantity x a The tracking error e of the reference model state variable is obtained by subtraction, and e is judged. If e is greater than or equal to a preset error threshold, the error is augmented to the state variable x. a The reference model's state tracking error e and regression vector Φ(x) are input to the adaptive controller and output by the adaptive controller u. ad For the main control output u main Compensation will be performed; otherwise, the main control output u will not be compensated. main Compensation is performed; the reference model of the model reference adaptive control loop is a closed-loop linear nominal control system based on the augmented linear quadratic regulating controller, and the adaptive controller of the model reference adaptive control loop is an e-corrected robust adaptive controller.

5. The multivariable robust adaptive control system for aero-engines as described in claim 4, characterized in that, The controller gain of the augmented linear quadratic regulator is optimized using a metaheuristic optimization method based on Newton-Raphson, with the goal of minimizing the tracking error.

6. The multivariable robust adaptive control system for aero-engines as described in claim 4, characterized in that, When switching from compensation to no compensation during dynamic compensation, the following method is used for perturbation-free switching: the integral of the output tracking error at time k is calculated as follows: yI Set to zero, and use u(k) = u0 + u main (k) represents the control variable for the aero-engine at time k, where u0 = u(k-1) = u main (k-1)+u ad (k-1) represents the control input of the aero-engine at the previous moment, u main (k) represents the main control output at time k.

Images