Aeroengine Transition State Control Method Based on Intelligent Model Predictive Control

Abstract

A transitional state control method for an aeroengine based on intelligent model predictive control belongs to the field of aeroengine control. First, the dynamic assignment method is used to improve the update strategies of position and velocity in the traditional CSO algorithm to achieve input constraint management, enabling the improved CSO algorithm to be used in MPC. Secondly, the backstepping method is used to improve the position update strategy in the traditional CSO algorithm to achieve output constraint management, enabling the improved CSO algorithm to be used in MPC. The improved CSO algorithm is combined with MPC to implement the intelligent model predictive control algorithm IMPC. Finally, the IMPC algorithm is applied to the transitional state control of the aeroengine. The present invention can directly select the controlled object as the prediction model, eliminating the possible errors between the prediction model and the controlled object. At the same time, the CSO algorithm is improved to achieve constraint management in IMPC, making the CSO algorithm easier to apply in practice and capable of improving the transitional state performance index under the condition of meeting all constraint conditions.

Description

Technical Field

[0001] The present invention belongs to the field of aero-engine control, and provides a method for controlling the transient state of an aero-engine based on intelligent model predictive control. Background Art

[0002] Model predictive control (MPC) is a control algorithm based on a predictive model. It derives the control action at each sampling time point by solving a finite-time interval constrained optimization problem. At a certain moment, by solving the optimization problem, an optimal control input sequence for a future period of time is obtained, and only the first value in the sequence is applied to the controlled object, and then this process is repeated at the next sampling instant.

[0003] Since MPC is a control algorithm based on a predictive model, it usually includes a predictive model to provide predictions for future periods. For a simple nonlinear device, the method of combining MPC with active disturbance rejection control (ADRC) is adopted in the literature [Yan Shuai, Liu Ziqi, Liu Haocheng, Cui Hanlin, Guan Haotian. Research and progress of particle swarm optimization algorithm [J]. Modern Industrial Economy and Informationization, 2019, 9(03): 19-20.]. First, an accurate mathematical model of the permanent magnet synchronous motor is established as the predictive model. Then, using polyhedral constraints and a quadratic cost function, a simple quadratic optimization problem is finally obtained and solved. However, for complex nonlinear devices such as aero-engines, it is difficult for existing research to describe complex nonlinear devices with a mathematical model. Therefore, there is inevitably an error between the predictive model and the controlled object, resulting in a decline in control performance. This prompts us to explore new MPC design methods to improve its efficiency and obtain better performance. In the present invention, it aims to propose a new MPC method that fully utilizes the information of the controlled object to provide predictions without having to establish an accurate mathematical prediction model.

[0004] In addition, some meta-heuristic algorithms achieve optimization by mimicking the behaviors of organisms in nature and do not always rely on mathematical models or mathematical derivations. Therefore, the combination of MPC and meta-heuristic algorithms may provide a possible approach to achieving our goal. The particle swarm optimization (PSO) algorithm, first proposed by Kennedy and Eberhart in [Yu Zhongqi, Du Zhaoping, Wang Weiran. Research on improved active disturbance rejection control system of permanent magnet synchronous motor based on model predictive control [J]. Transducers and Microsystem Technologies, 2022, 41(07): 52-56.], which is based on the idea of simulating the social behaviors of social animals such as fish schools and bird flocks, provides an effective method for dealing with optimization problems. This algorithm has advantages such as good optimization performance and easy implementation. However, one problem with the PSO algorithm is its poor performance when solving complex optimization problems with a large number of local optima. Then, some scholars proposed a particle swarm optimization algorithm integrating the idea of competition - the competitive swarm optimizer (CSO). CSO introduces a pairwise competition mechanism. The particles that win the competition will be directly passed to the next generation, while the position and velocity of the losers in the competition will be updated by learning from the winning particles before being passed to the next generation. In [Xie Jinfa, Liu Han, Li Bochao. PMSM vector control based on improved particle swarm optimization algorithm [J]. Modern Manufacturing Engineering, 2019, (07): 6-11.], the CSO algorithm is used to improve the optimization speed by using dynamic adaptive weights, an online optimization PSO-PID controller is established and applied to the control of permanent magnet synchronous motors. However, aiming at the defects of the single particle swarm optimization algorithm, such as being prone to falling into local minima, unable to limit constraint management, and difficult to be applied to MPC. The present invention proposes an improved particle swarm optimization algorithm (CSO), which changes the update strategy of particle positions by using the dynamic assignment method and the regression method, speeds up the convergence speed of particles, has higher accuracy and realizes limit constraint management, making it conducive to application in MPC. Then, the improved CSO algorithm is combined with MPC to propose a more effective intelligent model predictive control algorithm (IMPC). Finally, the designed IMPC is applied to a complex nonlinear system - an aeroengine, and the effectiveness of the IMPC algorithm in transient control is verified by simulation, improving the transient performance index.

[0005] Based on the above discussion, this patent studies a model predictive control method based on the improved CSO algorithm and proposes a new model predictive control algorithm - intelligent model predictive control (IMPC). Different from traditional MPC, IMPC fully utilizes the information of the controlled object to provide predictions without the need for an accurate mathematical prediction model. Then, taking the transient control of an aeroengine as an example, the effectiveness of the designed IMPC algorithm is verified, and the transient performance index is improved using the designed method under the condition of satisfying all constraint conditions. Summary of the Invention

[0006] Aiming at the problems existing in the prior art, a new MPC method is proposed in the present invention. This method first improves the traditional CSO algorithm. By using the dynamic assignment method and the retreat method to change the position and velocity update strategy of the particles, the learning ability of the particles is improved, the convergence speed is faster, the constraint management is realized, and it can be applied to MPC. Then, MPC is combined with the improved CSO algorithm to propose a new IMPC algorithm. This algorithm fully utilizes the information of the controlled object to provide predictions without having to establish an accurate mathematical model. Finally, the IMPC algorithm is applied to the transient control of aeroengines, and the effectiveness of the designed IMPC algorithm is verified through simulation. To achieve the above object, the technical solution adopted by the present invention is as follows:

[0007] An aeroengine transient control method based on intelligent model predictive control, comprising the following steps:

[0008] In the first step, the dynamic assignment method is used to improve the position and velocity update strategy in the traditional CSO algorithm, realize the constraint management of the input, and enable the improved CSO algorithm to be used in MPC.

[0009] The specific dynamic assignment method is as follows:

[0010] (1) Before using the CSO algorithm to solve the optimization problem, first determine the position and velocity of the particles. Let S(k) represent the swarm of m particles, where k represents the generation index. Each particle in the swarm has an n y +1-dimensional position and an n y +1-dimensional velocity. In the k-th generation, the position and velocity of the i-th particle are:

[0011]

[0012]

[0013] where i ∈ {1, 2,..., m}, the position of the particle represents different feasible control inputs of the optimization problem, and the position of each particle is X in (k) = u(t + (n - 1)h).

[0014] (2) Represent the position of the particle in step (1) with a one-dimensional vector. Assume the initial particle X(0) = (u0(t), u0(t + h),..., u0(t + n y h)). Then randomly initialize the first value of the position vector u0(t) with the upper and lower bounds, and the upper and lower bounds are updated by the following strategies respectively:

[0015] ub i = min(U max , u0(t + (i - 1)h) + ΔUmax ), (3)

[0016] lb i = max(U min , u0(t + (i - 1)h) + ΔU min ), (4)

[0017] (3) After setting the initial particles in step (2), according to the initial positions, calculate the fitness of the particles, that is, J i , J i The smaller the value of, the better the fitness of the particle. Particles with better fitness will be passed to the next generation S(k + 1) without update, while those with poor fitness should learn from those with good fitness and update their positions and velocities. Consider a particle X(k) = (u k (t), u k (t + h),..., u k (t + n y h)) in the k-th generation S(k) with velocity V(k) = (ΔU k (t), ΔU k (t + h), …, ΔU k (t + n y h)). If this particle loses the competition, then update the velocity of the particle according to the following learning strategy:

[0018]

[0019] where the subscripts l and w represent the information of the loser and the winner respectively, is the average position of the relevant particles, and R1, R2, R3 ∈ [0, 1] are randomly generated constants. The new velocity is expressed as

[0020]

[0021] (4) Check to ensure that the input constraint conditions in the optimization problem are satisfied, and let the corrected velocity satisfy any of the following constraint conditions: or where i = {1, 2, …, n y}}, and the corrected velocity V(k + 1) = (Δu k+1 (t), Δu k+1 (t + h),..., Δu k+1 (t + n y h)). Then the position of the particle can be updated with the corrected velocity V(k + 1) using the following strategy:

[0022]

[0023] (5) Use the dynamic assignment method to correct the new position, and the new position should also be checked and corrected to ensure that it does not exceed the limit. The new position before correction can be expressed as Then, using a strategy similar to that in (3) and (4), the new position X * (k + 1) can be derived, and finally through or correct the position. By this method, constraints (2) and (3) are satisfied during the optimization process, and finally the constraint management of the input is achieved.

[0024] In the second step, use the backtracking method to improve the position update strategy in the traditional CSO algorithm to achieve the constraint management of the output, so that the improved CSO algorithm can be used in MPC. Finally, combine the improved CSO algorithm with MPC to achieve the intelligent model predictive control algorithm (IMPC).

[0025] The specific backtracking method is as follows:

[0026] (1) Define the judgment function:

[0027]

[0028]

[0029]

[0030] Let L be the overrun index to describe whether the output exceeds the limit. Therefore, after initializing or updating the particle position, it is necessary to calculate the overrun index of the new particle to check whether the output exceeds the limit.

[0031] (2) Calculate L from step (1). If L > 0, it means that the output exceeds the constraint and the new particle is infeasible. To solve this problem, a method of backtracking to search for a new feasible particle is adopted. If the overrun index L of X(k + 1) is zero, then let the particle X(k) move to the new position X(k + 1) with the velocity V(k + 1). If the overrun index of X(k + 1) is not zero, then let the particle X(k + 1) return with the velocity -cV(k + 1), where c ∈ (0, 1) is a constant.

[0032] (3) Using the method of backtracking to search for feasible particles in step (2), the actual position update strategy is:

[0033]

[0034] where i = 1, 2, 3…, and then calculate again the overrun index of. If the index is zero, then this particle is feasible and can be passed to the next generation.

[0035] (4) Similarly, if the overrun index is not zero, it will return at a speed of -cV(k + 1) until the overrun index becomes zero. Therefore, using the backstepping method, the actual update strategy for the position is:

[0036]

[0037] In the formula If no feasible particle can be found using the backstepping method, it means the current speed is not feasible. Use formula (5) to reallocate R1, R2, and R3 to find a new feasible speed and ensure a feasible particle is found. Finally, during the optimization process of the CSO algorithm, the output constraint conditions in the transient state optimization objective function are satisfied.

[0038] (5) Since the improved CSO can achieve constraint management, combine the improved CSO algorithm with MPC to propose a new intelligent model predictive algorithm (IMPC). This algorithm fully utilizes the information of the controlled object to provide predictions without the need for an accurate mathematical prediction model. Using the prediction model, the optimal control input over a future time domain can be obtained by solving the optimization problem, making it more suitable for complex nonlinear systems such as aeroengines.

[0039] The third step is to apply the IMPC algorithm designed in the second step to the transient state control of an aeroengine.

[0040] (1) Obtain the output parameters of the aeroengine. First, select a certain engine dynamic simulation model, set different input signals in the input module to obtain the pre-designed fuel-air ratio FAR, and then obtain the output parameters of the engine at a certain operating state from the engine dynamic simulation model, including the rotational speeds of the low-pressure shaft (turbine) and high-pressure shaft (turbine), the low-pressure turbine outlet temperature, the surge margin, the pressure, etc., which can be exported from the output module.

[0041] (2) Then, obtain the aeroengine limit conditions. Determine the constraint conditions for the transient state control optimization problem of the engine according to the parameter limit conditions of the engine model, including the limit of the physical rotational speed N2 of the high-pressure compressor, the rotational speed N l limit, the limit of the high-pressure turbine outlet temperature T, the fan surge margin SM F boundary, the compressor surge margin SM C boundary, the fuel flow rate W f limit, and the transient state fuel flow rate rate ΔW f limit conditions. And build the corresponding constraint function according to the limit conditions, and its form is as follows:

[0042] N2 ≤ N 2,max (11)

[0043] Nl ≤ N l,max (12)

[0044] T ≤ T max (13)

[0045] SM C ≥ SM C,min (14)

[0046] W f,min ≤ W f ≤ ΔW f,max (15)

[0047] ΔW f ≤ ΔW f,max (16)

[0048] (3) Obtain the constraint conditions and output parameters of the aero - engine, as well as the designed IMPC algorithm from step (2). The IMPC algorithm can achieve constraint management and does not require an accurate mathematical model, thereby determining the optimization objective function for the engine transient control process, and realizing that during the transient process, the control input fuel - air ratio FAR does not exceed the high - pressure turbine outlet temperature T and the fan surge margin SM F boundary and other maximum limits, and control the limited output target speed N l within its constraint range to improve the transient control performance index.

[0049] (4) Verify the effectiveness of the improved CSO algorithm in the transient optimization process. First, represent the position of the particle as an input FAR sequence, and then, according to the intelligent model predictive control method designed in the second step, first, set the prediction time domain and the number of particles in the swarm, and set the terminal condition, i.e., the maximum number of iterations. Plot the output curves of multiple optimization processes. For each curve, it represents the change in the current global optimal fitness during the optimization process. During the optimization process, if the fitness for each curve is decreasing, it can be determined that the improved CSO algorithm is effective throughout the optimization process.

[0050] (5) Through simulation experiments, apply the designed IMPC algorithm to the aero - engine transient control, and obtain the change curves of the control input FAR and the high - pressure turbine outlet temperature T. If at the beginning of the transient process, the control input FAR gradually increases until the high - pressure turbine T reaches the constrained temperature, in order to ensure that T does not exceed the limit, the control input FAR slowly decreases during the intermediate transient process, and the fan surge margin SM F decreases, while the low - pressure turbine speed N l reaches the target speed, then it can be verified that throughout the transient process, the control input FAR does not exceed its maximum upper limit, and the output target speed N lThe transient performance index is improved within its limited transportation range.

[0051] In the fourth step, to verify the effectiveness of the present invention, the traditional CSO algorithm is applied to the transient control of an aeroengine, and the effectiveness of the improved IMPC algorithm is verified by comparison.

[0052] (1) Represent the change of control input during the entire transition process with particles. Since the traditional CSO cannot implement constraint management, without considering the constraints of the output, set the number of swarm particles and the number of iterations to obtain the change curves of the output control input FAR and the high-pressure turbine outlet temperature T.

[0053] (2) If the change curve obtained in (1) indicates that the traditional CSO algorithm cannot stably accelerate the control output N l to the target speed, it means that when the algorithm is used for the entire transient process, the dimension of a single particle is very high, becoming a large-scale optimization problem. Although a large number of particles and more iterations are set, the optimization ability of CSO is reduced, resulting in an unsatisfactory optimization result.

[0054] (3) Compared with the IMPC algorithm designed in the second step, the particles in the IMPC algorithm represent the change of control input within the prediction range, thus improving the optimization efficiency while reducing the number of particles and the number of iterations. In addition, compared with the traditional CSO algorithm, the biggest advantage of the IMPC algorithm is that it can implement constraint management, which makes our method more suitable for practical applications.

[0055] The beneficial effects of the present invention are as follows: The present invention proposes a new MPC algorithm, namely the IMPC algorithm. First, by using the dynamic assignment method and the step-back method, the position update strategy of the traditional CSO algorithm is improved to enable it to implement restricted constraint management and be applicable to MPC. Then, MPC is combined with the improved CSO algorithm to implement the IMPC algorithm. This algorithm can directly select the controlled object as the prediction model, replacing the commonly used mathematical prediction model in the traditional prediction model, eliminating the possible error between the prediction model and the controlled object. At the same time, the CSO algorithm is improved to implement the constraint management in IMPC, making the CSO algorithm easier to be applied in practice. In addition, the application of this method in the transient control of an aeroengine is studied. Using the designed IMPC algorithm, the transient performance index is improved under the condition of meeting all constraint conditions. Description of the Drawings

[0056] Figure 1 It is a schematic diagram of optimizing the CSO algorithm based on the dynamic assignment method;

[0057] Figure 2 It is a schematic diagram of optimizing the CSO algorithm based on the step-back method;

[0058] Figure 3 It is a schematic structural diagram of a JT9D engine model; in the figure, 1 is a fan, 2 is a low-pressure compressor, 3 is a high-pressure compressor, 4 is a combustion chamber, 5 is a high-pressure turbine, 6 is a low-pressure turbine, and 7 is a tail nozzle.

[0059] Figure 4 It is a schematic diagram of the simulation of a JT9D dynamic gas turbine;

[0060] Figure 5(a) shows the optimization process of the improved CSO algorithm; Figure 5(b) is an enlarged view of part A in Figure 5(a);

[0061] Figure 6 is a simulation diagram of the improved CSO algorithm applied to the transitional state control of an aeroengine; Figure 6(a) is a schematic diagram of the change in the control input FAR; Figure 6(b) is l a schematic diagram of the change in the output target speed N; Figure 6(c) is an enlarged view of part B in Figure 6(b);

[0062] Figure 7 It is a simulation diagram of the traditional CSO algorithm applied to the transitional state control of an aeroengine; Figure 7 (a) is a diagram of the change in the control input FAR; Figure 7 (b) is a schematic diagram of the change in the output target speed N l ; Specific implementation manners

[0063] The following further describes the present invention with reference to specific embodiments.

[0064] A method for transitional state control of an aeroengine based on intelligent model predictive control provided by an embodiment of the present invention includes the following steps:

[0065] In the first step, as Figure 1 , the update strategies of the position and speed in the traditional CSO algorithm are improved by using the dynamic assignment method to implement the constraint management of the input, so that the improved CSO algorithm can be used in MPC

[0066] 1.1 Determine the optimization function with input and output constraint conditions. First, consider the following optimization problem with constraints:

[0067]

[0068] s.t.U min ≤u(t + ih)≤U max , (2)

[0069] ΔU min ≤Δu(t + ih)≤ΔU max , (3)

[0070] Y min≤y(t + ih)≤Y max , (4)

[0071] where g is the objective function of the output y and the input u, c is the sampling step, and n y is the prediction horizon, U min , U max and ΔU min , ΔU max represent the minimum and maximum limits of the control input u and its change Δu respectively, Y min and Y max represent the minimum and maximum limits of the output, and t + ih represents the prediction of the relevant value i steps after the current time t.

[0072] 1.2 To solve the optimization problem in 1.1 using the CSO algorithm, let S(k) represent the swarm of m particles, where k represents the generation index. Each particle in the swarm has an n y +1 - dimensional position and an n y +1 - dimensional velocity. For example, in the k - th generation, the position and velocity of the i - th particle are:

[0073]

[0074]

[0075] where i ∈ {1, 2, …, m}, the position of the particle represents different feasible control inputs for the minimization problem (1)-(4), and the position of each particle is X in (k) = u(t+(n - 1)h). When solving the optimization problem, the position and velocity of the particle should be determined first. Due to the constraints on u and Δu, the initial position of the particle should be well - defined to satisfy the input constraint conditions (2) and (3) in the optimization function.

[0076] 1.3 Use a one - dimensional vector to represent the position of the initial particle in 1.2. Assume the initial particle X(0) = (u0(t), u0(t + h), …, u0(t + n y h)). Then randomly initialize the first value of the position vector u0(t) with upper and lower bounds, and the upper and lower bounds are updated by the following strategies respectively:

[0077] ub i = min(U max , u0(t+(i - 1)h)+ΔU max ), (7)

[0078] lb i = max(U min , u0(t+(i - 1)h)+ΔU min ), (8)

[0079] 1.4 After setting the initial particles, calculate the fitness of the particles according to the initial positions, i.e., J i . J i The smaller the value of J, the better the fitness of the particle. Particles with better fitness will be passed to the next generation S(k + 1) without update, while those with poor fitness should learn from those with good fitness and update their positions and velocities. Consider a particle X(k) = (u k (t), u k (t + h),..., u k (t + n y h)) in the k-th generation S(k) with velocity V(k) = (ΔU k (t), ΔU k (t + h), …, ΔU k (t + n y h)). Suppose this particle loses the competition, then update the velocity of the particle according to the following learning strategy:

[0080]

[0081] where the subscripts l and w represent the information of the loser and the winner respectively, is the average position of the relevant particles,

[0082] R1, R2, R3 ∈ [0, 1] are randomly generated constants. The new velocity is expressed as

[0083]

[0084] 1.5 After updating the velocity of the particle, it should be checked to ensure that the constraint (4) is satisfied, and let the corrected velocity satisfy any of the following constraint conditions or where i = {1, 2, …, n y}, and the corrected velocity V(k + 1) = (Δu k+1 (t), Δu k+1 (t + h),..., Δu k+1 (t + n y h)). Then the position of the particle can be updated with the corrected velocity V(k + 1) using the following strategy:

[0085]

[0086] 1.6 Use the idea similar to the dynamic assignment method (as shown in Figure 6) to check and correct the new position to ensure that it does not exceed the limit. The new position before correction can be expressed as Then, using the strategies similar to (7) and (8), the new position X* (k + 1)'s dynamic bound, and finally through or the correction position. By this method, constraints (2) and (3) are satisfied during the optimization process.

[0087] The second step is as Figure 2 , using the backstepping method to improve the position update strategy in the traditional CSO algorithm, realizing the constraint management of the output, enabling the improved CSO algorithm to be used in MPC, and finally combining the improved CSO algorithm with MPC to realize the intelligent model predictive control algorithm (IMPC).

[0088] 2.1 Define the judgment function:

[0089]

[0090]

[0091]

[0092] Let L be the overrun index to describe whether the output exceeds the limit. Therefore, after initializing or updating the particle position, it is necessary to calculate the overrun index of the new particle and check whether the output exceeds the limit.

[0093] 2.2 Calculate L from the formula of the judgment function in (1). If L > 0, it means the output exceeds the constraint and the new particle is infeasible. To solve this problem, the method of backstepping search for a new feasible particle is adopted. As Figure 2 , the arrow from X(k) to X(k + 1) represents the velocity V(k + 1). If the overrun index of X(k + 1) is not zero, then let the particle X(k + 1) return with the velocity -cV(k + 1), where c ∈ (0, 1) is a constant, and the actual position update strategy is:

[0094]

[0095] where i = 1, 2, 3…, and then calculate again the overrun index of, if the index is zero, then this particle is feasible and can be passed to the next generation.

[0096] 2.3 Similarly, if the overrun index of is not zero, it will adopt the method of backstepping search for a new feasible particle and return with the velocity -cV(k + 1) until the overrun index is zero. The actual position update strategy is:

[0097]

[0098] In the formula It can be found that the smaller the value of c, the stronger the search ability of the backtracking method. If no feasible particle can be found using the backtracking method, it means that the current velocity is not feasible. Then we can use formula (9) to reassign R1, R2, and R3 to find a new feasible velocity and ensure that a feasible particle is found. Finally, during the optimization process of the CSO algorithm, the output constraint condition (4) in the optimization function is satisfied.

[0099] 2.4 Since the improved CSO algorithm can achieve constraint management and does not require an accurate mathematical model, while traditional MPC must rely on an accurate mathematical model, the improved CSO algorithm is combined with MPC to propose a new intelligent model predictive algorithm (IMPC). This algorithm fully utilizes the information of the controlled object to provide predictions, does not require an accurate mathematical prediction model, and can obtain the optimal control input over a future time domain by solving the optimization problem, making it more suitable for complex nonlinear systems such as aeroengines.

[0100] In the third step, apply the IMPC method designed in the second step to the transient control of the aeroengine.

[0101] 3.1 Obtain the output parameters of the JT9D engine. First, select the JT9D engine. Its Simulink model is as Figure 3 shown. The schematic diagram of the JT9D engine model structure. In the figure, 1 is the fan, 2 is the low-pressure compressor, 3 is the high-pressure compressor, 4 is the combustion chamber, 5 is the high-pressure turbine, 6 is the low-pressure turbine, and 7 is the tail pipe.

[0102] As Figure 4 is the schematic diagram of the JT9D dynamic gas turbine engine simulation. The input of this model is the fuel-air ratio FAR. Different input signals are set in the input module to obtain the pre-designed fuel-air ratio. The JT9D engine model is an internal circulation system for calculating the internal parameters of the engine. After receiving the input signal, the rotor integrator integrates the derivatives of the two-shaft speeds calculated by the inner-loop device to obtain the speeds of the low-pressure shaft (turbine) and the high-pressure shaft (turbine) respectively. Other data, such as temperature, surge margin, pressure, etc., can be exported from the output module.

[0103] 3.2 Obtain the limitation conditions of the JT9D engine. Determine the constraint conditions for optimizing the transient control law of the engine according to the regulations of the engine dynamic simulation limitation conditions, including the limitation of the physical speed N2 of the high-pressure compressor, the limitation of the low-pressure turbine speed N l limitation, the limitation of the high-pressure turbine outlet temperature T, the surge margin SM F boundary of the fan, the surge margin SM C boundary of the compressor, the fuel flow rate W f limitation, and the transient fuel flow rate ΔW fConstraints. Then, based on the constraints of the JT9D engine, the constraint function is determined as

[0104] SM F ≥ 17%, 0.01 ≤ FAR ≤ 0.025, -0.001 ≤ ΔFAR ≤ 0.001, T ≤ 2100R.

[0105] 3.3 Apply the designed IMPC method to the transient control of aero - engines according to the constraints obtained in step (2). Since the designed IMPC method can satisfy the constraint management of input and output, we consider the following optimization problem to achieve that during the transient process, the control input fuel - air ratio FAR does not exceed its maximum limit, the limited output T is controlled within its constraint range, and the low - pressure turbine speed N l = 3867 r / min accelerates to another steady - state point as the initial point. In this case, due to the lack of a mathematical prediction model, SQP or other methods are ineffective.

[0106]

[0107] s.t. 0.01 ≤ FAR(t + ih) ≤ 0.025, (18)

[0108] -0.001 ≤ ΔFAR(t + ih) ≤ 0.001, (19)

[0109] T(t + ih) ≤ 2100R, (20)

[0110] 17% ≤ SM F (t + ih) (21)

[0111] 3.4 Verify the effectiveness of the improved CSO algorithm in the entire transient optimization process. The position of the particle is represented as a sequence of input FAR. Then, according to the intelligent model predictive control method designed in Section II, first, we set the prediction horizon n y = 1, the number of particles in the swarm num = 50, and set the terminal condition, that is, the maximum number of iterations i = 80. As shown in Fig. 5(a), for illustration purposes, the first five optimization processes are plotted.

[0112] The horizontal axis is the number of iterations, and the vertical axis is the current global - best fitness. For each curve, it represents the change of the current global - best fitness during the optimization process. It can be seen that during the optimization process, the fitness is decreasing. Therefore, CSO has a certain exploration ability during the optimization process. During the entire optimization process, the fitness also decreases (from about 179 to 165, different curves are shown in Fig. 5(b)). This means that the improved CSO algorithm is effective during the entire optimization process.

[0113] 3.5 Through simulation experiments, the designed IMPC algorithm is applied to the transient control of aeroengines. The final control results are shown in Figure 6. The control input FAR is shown in Figure 6(a), where the lower broken line represents the change of FAR, and the upper dotted line represents the upper limit of FAR. At the beginning of the transient process, the control input FAR increases with its maximum growth until the high-pressure turbine T reaches the constrained temperature. To ensure that T does not exceed the limit, the control input FAR slowly decreases during the intermediate transient process, and at the same time, the surge margin SM of the fan F decreases. As shown in Figure 6(b) and Figure 6(c), at the same time, the low-pressure turbine speed N l steadily rises from 3688 r / min to 3867 r / min. Although there is an overshoot around 1 s, the overshoot of N l is less than 0.2%, which is within a reasonable range. During the entire transient process, the control input does not exceed its upper limit, and the output is within its constraint range. Therefore, the designed IMPC algorithm is effective.

[0114] Fourthly, in order to verify the effectiveness of the present invention, the traditional CSO algorithm is applied to the transient control of aeroengines, and the effectiveness of the improved IMPC algorithm is verified by comparison.

[0115] 4.1 Use particles to represent the change of the control input during the entire transition process. Since the traditional CSO algorithm cannot implement constraint management, the constraints of the output are not considered. Let the number of swarm particles num = 300000 and the number of iterations i = 100000. As Figure 7 (a) is the curve of the output control input FAR, and as Figure 7 (b) is the curve of the change of the high-pressure turbine outlet temperature T.

[0116] 4.2 From the change curves of FAR and T obtained in 4.1, it can be found that the traditional CSO algorithm cannot stably accelerate the control output N l to 3867 r / min. This is because the algorithm is used for the entire transient process. The dimension of a single particle is very high, becoming a large-scale optimization problem. In this case, although we set a large number of particles and more iterations, the optimization ability of CSO is reduced, resulting in an unsatisfactory optimization result. In the improved CSO algorithm, the particles represent the change of the control input within the prediction range, thus improving the optimization efficiency with fewer particles and iterations. In addition, compared with the traditional CSO algorithm, the biggest advantage of this algorithm is that it can implement constraint management, which makes our method more suitable for practical applications.

[0117] This patent proposes a new MPC algorithm, namely the IMPC algorithm, which uses an improved CSO algorithm in the optimization process. This model directly selects the controlled object as the prediction model, replacing the commonly used mathematical prediction model in traditional prediction models, and eliminating the possible errors between the prediction model and the controlled object. At the same time, the improved CSO algorithm realizes the constraint management in IMPC, making the CSO algorithm more easily applied in practice. In addition, the application of this method in the transitional state control of aeroengines is studied, and using the designed method, the transitional state control goal is achieved under the condition of meeting all constraint conditions.

[0118] The embodiments described above only represent the implementation manners of the present invention, but should not be construed as limiting the scope of the invention patent of the present invention. It should be noted that for those skilled in the art, without departing from the concept of the present invention, several deformations and improvements can still be made, and these all belong to the protection scope of the present invention.

Claims

1. A transitional state control method for an aeroengine based on intelligent model predictive control, characterized in that, It includes the following steps: In the first step, the dynamic assignment method is used to improve the update strategy of position and velocity in the traditional CSO algorithm, realize the constraint management of the input, and enable the improved CSO algorithm to be used in MPC; In the second step, the regression method is used to improve the position update strategy in the traditional CSO algorithm, realize the constraint management of the output, enable the improved CSO algorithm to be used in MPC, and finally combine the improved CSO algorithm with MPC to realize the intelligent model predictive control algorithm IMPC; the CSO algorithm is the competitive swarm optimizer; In the third step, the IMPC algorithm designed in the second step is applied to the transient control of aeroengines: (1) Obtain the output parameters of the aeroengine; First, select a certain engine dynamic simulation model, set different input signals in the input module to obtain the pre-designed fuel-air ratio FAR, and then obtain the output parameters of the engine at a certain working state from the engine dynamic simulation model, including the rotational speeds of the low-pressure shaft and the high-pressure shaft, the low-pressure turbine outlet temperature, the surge margin, and the pressure data, which are exported from the output module; (2) Obtain the limitation conditions of the aero-engine, and determine the constraint conditions of the engine transient control optimization problem according to the regulations of the limitation conditions of each parameter of the engine model, including the limitation of the physical speed N2 of the high-pressure compressor, the speed N l limitation, the limitation of the high-pressure turbine outlet temperature T, the fan surge margin SM F boundary, the compressor surge margin SM C boundary, the fuel flow rate W f limitation and the transient fuel flow rate ΔW f limitation conditions; and establish the corresponding constraint function according to the limitation conditions, and its form is as follows: N2 ≤ N 2,max (11) N l ≤ N l,max (12) T≤T max (13) SM C ≥SM C,min (14) W f,min ≤W f ≤ΔW f,max (15) ΔW f ≤ΔW f,max (16) (3) Obtain the constraint conditions and output parameters of the aero-engine, as well as the designed IMPC algorithm from step (2). The IMPC algorithm can achieve constraint management and does not require an accurate mathematical model, thereby determining the optimization objective function for the engine transient control process, and realizing that during the transient process, the controlled input fuel-air ratio FAR does not exceed the high-pressure turbine outlet temperature T and the fan surge margin SM F Boundary maximum limits such as etc., and the limited output target speed N l is controlled within its constraint range to improve the transient control performance index; (4) Through simulation experiments, apply the designed IMPC algorithm to the transient control of aeroengines to obtain the variation curves of the control input FAR and the high-pressure turbine outlet temperature T; if at the beginning of the transient process, the control input FAR gradually increases until the high-pressure turbine T reaches the constrained temperature, in order to ensure that T does not exceed the limit, the control input FAR slowly decreases during the intermediate transient process, and the surge margin SM of the fan F decreases, while the low-pressure turbine speed N l reaches the target speed, then it can be verified that during the entire transient process, the control input FAR does not exceed its maximum upper limit, and the output target speed N l within the transportation range of its limiting conditions, improve the transient performance index.

2. The transitional state control method for an aeroengine based on intelligent model predictive control according to claim 1, characterized in that, The specific dynamic assignment method in the first step is as follows: (1) Before using the CSO algorithm to solve the optimization problem, first determine the position and velocity of the particles. Let S(k) represent the bee colony of m particles, where k represents the generation index; each particle in the colony has an (n y + 1)-dimensional position and an (n y + 1)-dimensional velocity. In the k-th generation, the position and velocity of the i-th particle are: where \(i\in\{1,2,\ldots,m\}\), the position of the particle represents different feasible control inputs for the optimization problem, and the position of each particle is \(X\) in (k)=u(t+(n - 1)h); (2) Represent the positions of the particles in step (1) using one-dimensional vectors; assume the initial particle X(0) = (u0(t), u0(t + h), …, u0(t + n y h)); then randomly initialize the first value of the position vector u0(t) with upper and lower bounds, and the upper and lower bounds are updated by the following strategies respectively: ub i = min(U max , u0(t + (i - 1)h) + ΔU max ), (3) lb i = max(U min , u0(t + (i - 1)h) + ΔU min ), (4) (3) After setting the initial particles in step (2), calculate the fitness of the particles based on their initial positions, i.e., J i ; Particles with good fitness will be passed to the next generation S(k + 1) without update, while those with poor fitness should learn from the ones with good fitness and update their positions and velocities. Consider a particle X(k) = (u k (t), u k (t + h),..., u k (t + n y h)) in the k-th generation S(k) with velocity V(k) = (ΔU k (t), ΔU k (t + h), …, ΔU k (t + n y h)). If this particle loses the competition, then update the particle's velocity according to the following learning strategy: where the subscripts l and w represent the information of the loser and the winner respectively, is the average position of the relevant particles, and R1, R2, R3 ∈ [0, 1] are randomly generated constants; the new velocity is expressed as: (4) Check to ensure that the input constraints in the optimization problem are satisfied, and let the corrected velocity satisfy any of the following constraints: or where i = {1, 2, …, n y}, and the corrected velocity V(k + 1) = (Δu k+1 (t), Δu k+1 (t + h),..., Δu k+1 (t + n y h)); then the position of the particle can be updated with the corrected velocity V(k + 1) using the following strategy: (5) Use the dynamic assignment method to correct the new position. The new position should also be checked and corrected to ensure that it does not exceed the limit; the new position before correction can be expressed as Then, using a strategy similar to that in (3) and (4), the new position X * (k + 1) of the dynamic bounds can be derived, and finally through or correct the position. By this method, the input constraint conditions in the transition state optimization objective function are satisfied during the optimization process, and finally the input constraint management is achieved.

3. The transitional state control method for an aeroengine based on intelligent model predictive control according to claim 1, characterized in that, The specific regression method in the second step is as follows: (1) Define a judgment function: Let L be the overrun index, which is used to describe whether the output exceeds the limit. Therefore, after initializing or updating the particle position, it is necessary to calculate the overrun index of the new particle to check whether the output exceeds the limit; (2) Calculate L from step (1). If L>0, it means that the output exceeds the constraint and the new particle is infeasible. To solve this problem, a method of backward searching for a new feasible particle is adopted; if the overrun index L of X(k + 1) is zero, then let the particle X(k) move to the new position X(k + 1) with the velocity V(k + 1). If the overrun index of X(k + 1) is not zero, then let the particle X(k + 1) return with the velocity -cV(k + 1), where c∈(0,1) is a constant; (3) Using the method of backward searching for feasible particles in step (2), the actual update strategy of the position is: where i = 1, 2, 3..., and then calculate again of the overrun index. If the index is zero, then this particle is feasible and can be passed to the next generation; (4) Similarly, if has a non-zero overrun index, it will return at a speed of -cV(k + 1) until the overrun index is zero; therefore, using the backstepping method, the actual update strategy for the position is: where If no feasible particle can be found using the backtracking method, it indicates that the current velocity is not feasible. Formula (5) is used to reallocate R1, R2, and R3 to find a new feasible velocity and ensure that a feasible particle is found. Finally, during the optimization process of the CSO algorithm, the output constraint conditions in the transition state optimization objective function are satisfied; (5) Since the improved CSO can realize constraint management, the improved CSO algorithm is combined with MPC to obtain the intelligent model prediction algorithm IMPC.

Images