CN110219736B - Direct thrust control method of aero-engine based on nonlinear model predictive control - Google Patents

Abstract

The invention discloses an aero-engine direct thrust control method based on nonlinear model predictive control. The method of the present invention directly takes the thrust as the control target, instead of the method in which the traditional control method adopts unmeasurable parameters as the control target. The online sliding window deep neural network is used as the prediction model. The model adopts the deep learning structure, which can improve the accuracy of the model, and select the training data online based on the sliding window method, which reduces the sensitivity to the noise of the training data. Compared with the current popular control methods, the proposed method shortens the acceleration time by 0.425 seconds and increases the response speed by about 1.14 times.

Description

Direct thrust control method of aero-engine based on nonlinear model predictive control

technical field

The invention relates to an aero-engine control method, in particular to an aero-engine direct thrust control method based on nonlinear model predictive control.

Background technique

The main function of an aeroengine is to provide thrust to an aircraft quickly and accurately. Traditional engine control strategies are sensor-based, ie indirectly control thrust by controlling engine measurable parameters such as rotor speed, engine pressure ratio (EPR), or other measurable parameters. However, due to factors such as degradation, manufacturing and manufacturing tolerances, the correspondence between engine thrust and measurable parameters changes. Therefore, if the traditional control idea is adopted, the engine thrust control error must exist. In addition, in order to ensure the safe and stable operation of the engine at the worst operating point, the traditional control strategy often maintains a large safety margin, which will greatly limit the performance of the engine at other operating points.

In view of these shortcomings, the researchers invented a new control strategy-Model-Based Engine Control (MBEC), which directly controls the engine performance. Among them, Model Predictive Control (MPC) is an important key technology and research field in model-based engine control. MPC has better acceleration performance than traditional controllers, and has aroused extensive research interest in the field of aero-engine control. VroemenBG et al. applied MPC technology to laboratory gas turbines. Brunell et al. studied the feasibility of nonlinear model predictive control (NMPC) with constraints for state estimation, and applied it to the simulation model of turbojet engine. DeCastro proposed an NMPC based on linearly variable parameters and applied it to the active tip clearance of a commercial turbofan engine. Richter proposed an implementation method of multiplexing, which greatly reduces the computational complexity of the NMPC optimization algorithm. Di Cairano et al. developed an MPC strategy to control engine speed to ground idle through electronic throttle and spark timing. The model prediction model of the above work mainly focuses on the linear model, and has achieved good control effects. However, the aero-engine is a strong nonlinear power plant, and the modeling error of the linear model is inevitable. Furthermore, the control objectives of conventional NMPC are measurable parameters. However, the corresponding relationship between the measurable parameters and the thrust or power of the engine changes with the service time of the engine, which leads to a decrease in the thrust control accuracy.

SUMMARY OF THE INVENTION

The technical problem to be solved by the present invention is to overcome the deficiencies of the prior art, and to provide a direct thrust control method of an aero-engine based on nonlinear model predictive control, which can effectively improve the response capability of the engine.

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

A direct thrust control method of aero-engine based on nonlinear model predictive control, which takes thrust as the direct control target, and uses nonlinear model predictive control method for control. Specifically, it is realized by solving the following rolling optimization problem:

为控制目标预测值，u为控制变量向量，N_f、N_c分别为风扇转速、压气机转子转速，S_mf、S_mc分别为风扇喘振裕度、压气机喘振裕度，T₄₁为高压涡轮进口温度，Q和R正定对称，正整数N_u和N_p分别为控制时域和预测时域。where r is the engine control command,

is the predicted value of the control target, u is the control variable vector, N _f , N _c are the fan speed and compressor rotor speed, respectively, S _mf , S _mc are the fan surge margin and the compressor surge margin, respectively, T ₄₁ is For the inlet temperature of the high _- pressure turbine, Q and R are positively definite symmetrical, and the positive integers Nu and N _p are the control time domain and the prediction time domain, respectively.

Preferably, a pre-trained online sliding window deep neural network is used as the prediction model of the nonlinear model predictive control method, and the loss function of the online sliding window deep neural network is described as:

where x is the input vector, y is the output vector, j is the jth moment, L is the length of the rolling interval, f _DNN represents the mapping of the deep neural network, W is the weight matrix, and b is the bias vector.

Further preferably, the input x _DNN and output y _DNN of the prediction model are as follows:

Among them, W _fb (k), N _f (k), N _c (k), S _mf (k), S _mc (k), T ₄₁ (k), F (k) are the engine fuel input at time k, respectively , fan speed, compressor rotor speed, fan surge margin, compressor surge margin, high pressure turbine inlet temperature, engine thrust, m ₁ , m ₂ , ..., m ₇ are preset positive integers.

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

The invention can significantly improve the control precision and response speed of the aero-engine control system.

Description of drawings

1 is a schematic structural diagram of a nonlinear model predictive control system of the present invention;

Figure 2 is a schematic diagram of the principle of rolling optimization;

Figure 3 is a deep neural network diagram;

Figure 4 is a schematic diagram of the sliding window principle;

Figure 5 is a schematic diagram of the principle of the back-propagation algorithm;

6(a) to 6(g) are simulation results of the method of the present invention, in which F, T ₄₁ , N _f , N _c , S _mf , and S _mc are normalized.

Detailed ways

In view of the deficiencies of the prior art, the idea of the present invention is to take thrust as the direct control target, and use the nonlinear model predictive control method for control, thereby improving the control precision and response speed of the aero-engine control system.

Below in conjunction with accompanying drawing, technical scheme of the present invention is described in detail:

The structure of the nonlinear model predictive control system of the present invention is shown in FIG. 1 . The engine nonlinear model estimates unmeasurable parameters. The NMP module includes an optimization sub-module and a prediction model sub-module. The present invention adopts an online sliding window deep neural network (OL-SW-DNN) as a prediction model, which is more efficient than traditional neural network and support vector regression. Other shallow network structures have strong fitting ability and improve the prediction accuracy. In the optimization sub-module, the optimization target of NMPC is the thrust command. It can be seen that the optimization sub-module and the prediction model sub-module of NMPC are the key technologies, as follows.

Nonlinear model predictive control is a closed-loop, constrained online optimization control. As shown in Figure 2, at time k, the control objective is:

Among them, r is the control command, In order to control the target predicted value, u is the control variable vector, and Q and R are positive definite symmetry. The positive integers _Nu and _Np are the control interval and the prediction interval, respectively.

The control objective of traditional aero-engines is always to select measurable parameters, such as engine boost ratio or rotor speed. But the relationship between thrust and measurable parameters will change over time of use. Therefore, the control objective of the present invention is to directly control the thrust F.

Engine operation is limited by mechanical and component temperature limitations, such as fan and compressor rotor speed limitations, fan and compressor surge margin limitations, and high-pressure turbine inlet temperature limitations. In addition, the physical constraints of the actuator should also be considered. Therefore, in order to ensure the safe and stable operation of the engine, the engine operation should meet the following constraints:

Among them, N _f and N _c are the fan and compressor rotor speeds, respectively, S _mf and S _mc are the fan and compressor surge margins, respectively, and T ₄₁ is the high-pressure turbine inlet temperature.

The control objectives and limit constraints in Equation (1) and Equation (2) are calculated by the prediction model. The accuracy of the prediction model will directly affect the control effect. The traditional NMPC prediction model is a linear model. However, the transient process of aero-engine is a strongly nonlinear process. Therefore, an NMPC based on OnLine Sliding Window Deep Neural Network (OL-SW-DNN) was invented as a prediction model.

DNN has strong fitting ability to nonlinear objects, which can be described as:

y=f _DNN (x) (3)

where x is the input vector and y is the output vector. In order to maintain the dynamic characteristics of the engine, the input of the prediction model includes the current and historical engine fuel input W _fb , and the historical moment S _mf , S _mc , N _f , T ₄₁ and F . The inputs and outputs of the predictive model are:

Among them, the selection of m ₁ , m ₂ ,..., m ₇ is related to the nonlinearity of the engine. As shown in Figure 3 and Figure 4, DNN has a multi-layer network structure. The more hidden layers of DNN, the stronger the fitting ability of DNN. Each layer of DNN can be defined as:

a ^l+1 =W ^l h ^l +b ^l (5)

h ^l+1 =σ(a ^l+1 ) (6)

where W ^l is the weight matrix, b ^l is the bias vector, σ is the activation function, h ^l (l>0) is the output of the l layer, l=1,2,...,n _l ,n _l is the output of each layer The number of hidden nodes.

The loss function of OL-SW-DNN is described as:

W and b are updated as follows:

in, respectively , and η is the learning rate.

和

可以通过图5所示的反向传播算法计算得到：

and

It can be calculated by the back-propagation algorithm shown in Figure 5:

where ^δl is:

是Hadamard积

Among them, l=n _net ,n _net -1,L,2,

is the Hadamard product

为：make

for:

where _nnet is the number of network layers.

To verify the effectiveness of the method, the NMPC based on the method of the present invention and the popular NMPC based on Extended Kalman Filter (EKF) are simulated respectively. The simulation process for both methods selects the engine acceleration process. The acceleration process takes the steady-state working point of the throttle lever angle PLA=26° as the starting point and the steady-state working point of PLA=70° as the end point. The operating conditions for both simulations are standard atmospheric conditions at altitude H = 0 km and Mach number Ma = 0. The accompanying figure shows the acceleration process simulation of EKF-based NMPC and the proposed NMPC, where 'NMPC-EKF' represents EFK-based NMPC, and 'NMPC-DNN' represents OL-SW-DNN-based NMPC. For convenience, these two methods are described below as conventional NMPC and proposed NMPC, respectively. The engine parameters in the figure are normalized.

As shown in Fig. 6(a), in the conventional NMPC and the proposed NMPC, the time for the engine thrust to increase to 100% thrust is 3.025 seconds and 2.6 seconds, respectively. Compared with the conventional NMPC, the proposed NMPC reduces the acceleration time by 0.425 seconds and increases the acceleration speed by almost a factor of 1.14. The main reason is that the traditional NMPC adopts a linear model as a prediction model. However, the transient process of the engine is a strongly nonlinear process. With linear models as prediction models, prediction errors are inevitable. The invented NMPC uses OL-SW-DNN as the prediction model. OL-SW-DNN has strong fitting ability for nonlinear objects, which improves the prediction accuracy of NMPC.

As shown in Fig. 6(g), the operating point of the engine during acceleration moves along the surge boundary, which is considered to be the path with the fastest engine acceleration response. For other constraints, as shown in Figures 6(c)-6(f), the engine did not reach overtemperature, overspeed, or surge. Therefore, the control method of the present invention has higher control precision and response speed.

Claims (3)

1. A direct thrust control method of an aircraft engine based on nonlinear model predictive control is characterized in that thrust is taken as a direct control target, and the control is carried out by using a nonlinear model predictive control method; specifically, the method is realized by solving the following rolling optimization problem:

wherein r is an engine control command,for control target prediction, u is the control variable vector, N_f、N_cRespectively the fan speed, the compressor rotor speed, S_mf、S_mcRespectively fan surge margin, compressor surge margin, T₄₁For high pressure turbine inlet temperature, Q and R are positively symmetrical, and N is a positive integer_uAnd N_pRespectively a control time domain and a prediction time domain.

2. The nonlinear model predictive control-based aircraft engine direct thrust control method according to claim 1, characterized in that a pre-trained online sliding window deep neural network is used as a predictive model of the nonlinear model predictive control method, and a loss function of the online sliding window deep neural network is described as:

where x is the input vector, y is the output vector, j is the jth time, L is the length of the rolling interval, f_DNNAnd (5) expressing the mapping of the deep neural network, wherein W is a weight matrix, and b is a bias vector.

3. The nonlinear model predictive control-based aircraft of claim 2Method for direct thrust control of an empty engine, characterized in that the input x of the prediction model is_DNNAnd output y_DNNThe method comprises the following specific steps:

wherein, W_fb(k)，N_f(k),N_c(k),S_mf(k),S_mc(k),T₄₁(k) F (k) engine fuel input, fan speed, compressor rotor speed, fan surge margin, compressor surge margin, high pressure turbine inlet temperature, engine thrust, m, at time k, respectively₁,m₂,…,m₇Is a preset positive integer.

Images