阅读说明：本技术 一种航空发动机推力衰退缓解的神经网络控制方法 (A kind of neural network control method that aeroengine thrust decline is alleviated ) 是由 鲁峰 闫召洪 黄金泉 仇小杰 于 2019-07-23 设计创作，主要内容包括：本发明公开了一种航空发动机推力衰退缓解的神经网络控制方法,该方法包括：采用神经网络学习的NARMA-L2模型建立内环控制系统；结合内环控制系统,设计基于变增量LP优化算法的外环指令修正模块,得到在发动机发生部件性能突变的情况下具有自调整能力的推力衰退缓解神经网络控制器。本发明解决了常规多变量控制器在发动机发生部件性能突变的情况下整机推力水平下降的问题,适用于在一定飞行包线内不同工作点的发动机推力衰退缓解控制,对于在不超温、不超转的情况下缓解由发动机部件性能突变引起的整机推力损失、提高发动机整机性能表现有着积极促进的作用。(The invention discloses the neural network control methods that a kind of decline of aeroengine thrust is alleviated, this method comprises: using the NARMA-L2 model foundation inner loop control system of neural network learning；It in conjunction with inner loop control system, designs and correction module is instructed based on the outer ring for becoming increment LP optimization algorithm, obtain the thrust decline in the case where engine generation part performance is mutated with self-adjusting ability and alleviate nerve network controller.The present invention solves the problems, such as that conventional multivariable controller complete machine thrust level in the case where the mutation of engine generation part performance declines, suitable for the motor power decline alleviation control of the different operating point in certain flight envelope, for alleviating the complete machine thrust loss caused by the mutation of engine components performance in the case where not overtemperature, not excess revolutions, raising engine overall performance performance plays the role of actively promoting.)

1. a kind of neural network control method that aeroengine thrust decline is alleviated, which comprises the following steps:

Step A) physical descriptor interpolation model between engine high pressure rotor speed, pressure ratio and jet pipe throat area is established, And using the relationship between main fuel amount and high pressure rotor revolving speed, the inner ring control alleviated using neural network performance degradation NARMA-L2 model processed；

Step B) NARMA-L2 model is combined, it designs and correction module is instructed based on the outer ring for becoming increment LP algorithm, exploitation is being started Nerve network controller is alleviated in the thrust decline with self-adjusting ability in the case that machine generation part performance is mutated；

Step C) engine generation part performance mutation when, by outer ring be added instruction correction module, to engine instruct carry out Amendment tracks revised instruction using thrust decline alleviation nerve network controller and then achievees the purpose that thrust is restored.

2. a kind of neural network control method that aeroengine thrust decline is alleviated according to claim 1, feature Be: the step A) in using the relationship between main fuel amount and high pressure rotor revolving speed, declined using neural network performance Moving back the inner loop control NARMA-L2 model of alleviation, specific step is as follows:

Step A1), choose engine main fuel amount as input quantity u [k], high pressure rotor revolving speed is as output quantity y [k], and root NARMA-L2 model is constructed according to selected input and output amount are as follows:

Y [k]=f₀(y[k-1],y[k-2],…,y[k-n],u[k-1],u[k-2],…,u[k-n])+g₀(y[k-1],y[k- 2],…,y[k-n],u[k-1],u[k-2],…,u[k-n])u[k]

Wherein, f₀And g₀It is two Nonlinear Mappings in MARMA-L2 model, passes through training neural network group f₀ ^*And g₀ ^*It is approximate It obtains, and revolving speed control law are as follows:

Wherein, y^*[k] is engine desired output；And according to the instruction of engine complete machine blow down ratio, reality output blow down ratio and output Revolving speed interpolation converts to obtain jet pipe throat area control amount；

Step A2), utilize the error e between desired output and reality output_c(k) to neural network group f₀ ^*And g₀ ^*Topological structure ginseng Number carries out on-line amending:

W (k+1)=W (k)+α Δ W (k)

V (k+1)=V (k)+α Δ V (k)

Wherein, W and V is respectively the weight and threshold value of neural network group；α is learning rate, and size affecting parameters are regulated the speed, By repeatedly trying to follow the example of determination；Δ W (k) and Δ V (k) is Jacobian matrix；Each element in Topology Vector the k moment update such as Under:

w_i(k+1)=w_i(k)-αΔw_i(k)

v_i(k+1)=v_i(k)-αΔv_i(k)

Wherein, T={ y (k), y (k-1), y (k-2) ..., y (k-n), u (k-1), u (k-2) ..., u (k-n) }.

3. a kind of neural network control method that aeroengine thrust decline is alleviated according to claim 1, feature It is: the step B) it is middle in conjunction with NARMA-L2 model, it designs and correction module is instructed based on the outer ring for becoming increment LP algorithm, obtain Nerve network controller tool is alleviated in thrust decline in the case where the mutation of engine generation part performance with self-adjusting ability Steps are as follows for body:

Step B1), it constructs and nerve network control system structure chart, packet is alleviated based on the aeroengine thrust for becoming increment LP optimization Include inner ring nerve network controller, engine mockup, performance state filter, outer ring instruction correction module；

Step B2), engine initial order is determined according to flying condition and relevant control plan, is exported using engine mockup Correction module is instructed based on the outer ring for becoming increment LP algorithm with restrictive condition design, establishes and is mutated in engine generation part performance In the case where with self-adjusting ability thrust decline alleviate nerve network controller.

4. a kind of neural network control method that aeroengine thrust decline is alleviated according to claim 3, feature It is: the step B2) it is repaired according to engine mockup output and restrictive condition design based on the outer ring instruction for becoming increment LP algorithm Positive module establishes the thrust decline in the case where mutation of engine generation part performance with self-adjusting ability and alleviates nerve net Specific step is as follows for network controller:

Step B2.1), to not measurable propulsive effort and boundary restrictive condition, the small range near operating point is linearized, and extracts performance Linear relationship PSM between index, amount of restraint and control amount, specific formula are as follows:

Wherein, Δ S_mfWith Δ S_mcThe respectively surge margin increment of fan and compressor, Δ n_LWith Δ n_HRespectively low pressure, high pressure Rotor speed increment, Δ F_nFor thrust increment, Δ T_t6For temperature increment after turbine, Δ W_fFor amount of fuel increment, Δ A₈For nozzle Area increase, P matrix element can be obtained from Stable status engine model by method of perturbation, specific formula are as follows:

P_ij={ M_i(u₀(j)+Δu_j,...)-M_i(u₀)}/Δu_j

I=1,2 ..., 6, j=1,2

Wherein, M_i() is the output matrix being calculated by Stable status engine model；u₀It is the initial value of control amount；Δu_j It is j-th of control amount increment；

Step B2.2), for the control amount increment Delta u in step B2.1)_j, can be by corresponding control amount u_jIt is obtained multiplied by increment coefficient Out, the selection rule of increment coefficient are as follows:

Wherein, k is the number of iterations；ΔF_nFor thrust increment；According to the PSM extracted in step B2.1), instruction amendment is converted to LP optimization problem solves the correction amount instructed needed for thrust is alleviated.

5. a kind of neural network control method that aeroengine thrust decline is alleviated according to claim 4, feature Be: the step B2.2) in instruction amendment is converted into LP optimization problem, solve the amendment measurer instructed needed for thrust is alleviated Steps are as follows for body:

Step B2.2.1), in engine not excess revolutions, overtemperature, surge margin and control amount do not meet area requirement, thrust is no more than Searching engine maximum thrust under the constraint of desired value, specific formula are as follows:

Wherein, F_nFor motor power, W_fFor amount of fuel, A₈For nozzle area, u=[W_f,A₈]^TFor engine dominant vector, u_minAnd u_maxRespectively vector lower bound and the upper bound, n_LminAnd n_LmaxRespectively rotational speed of lower pressure turbine rotor is minimum and peak, n_HminWith n_HmaxRespectively high pressure rotor revolving speed is minimum and peak, T_t6minAnd T_t6maxTemperature is minimum respectively after turbine and peak, S_mf And S_mcRespectively fan and compressor surge nargin, F_norThe motor power for the failure that do not mutate is calculated for real-time model Desired value；

Step B2.2.2), according to control quantity constraint and restriction, using the PSM extracted in step B2.1), by step B2.2.1 the formula in) is rewritten are as follows:

Similarly acquire maximum thrust performance indicator are as follows:

ΔF_n=p₆₁ΔW_f+p₆₂ΔA₈

Step B2.2.3), LP problem is solved, is iterated to after meeting required precision or reach maximum number of iterations limitation To globally optimal solution W_fAnd A₈, and optimal solution is input in engine, corresponding output signal n is calculated_HAnd EPR, by two Command signal makees poor, obtained Δ n before person and thrust are restored_HrWith Δ EPR_rIt is repaired as what rotary speed instruction and complete machine blow down ratio instructed Positive value is input to inner ring nerve network controller, and it is whole to alleviate engine for the instruction after inner ring nerve network controller tracking correction Machine thrust loss.

6. a kind of neural network control method that aeroengine thrust decline is alleviated according to claim 1, feature It is: step C) in, when engine generation part performance is mutated, first by engine speed instruction decline 1%, outer ring is then added Repair instruction.

Technical field

Alleviate control technology field the invention belongs to aeroengine thrust more particularly to a kind of aeroengine thrust declines Move back the neural network control method of alleviation.

Background technique

With the rapid development of modern technical aeronautics, aircraft needs that performance is stronger, the higher aero-engine of reliability. As the structure of engine complexity itself and working environment severe locating for it, after engine runs a period of time, engine All parts different degrees of performance mutation can occur due to burn into foreign object damage etc..These reasons make practical hair Motivation and with that can generate deviation between model rated engine, is mainly reflected in the reduction of the negotiability and efficiency of component.For Safe flight, the increasing service life of engine for ensuring aircraft need to carry out the engine that generation part performance is mutated Thrust is restored, and therefore, it is very necessary that control method is alleviated in research aeroengine thrust decline.

A kind of effective thrust, which alleviates control method, and then to be reached and pushes away by inner loop control revolving speed, outer ring revision directive The purpose that power is alleviated.Aero-engine is a complicated thermodynamic system, has very strong uncertain and time variation, this makes The design for obtaining inner loop control device becomes difficult.Neural network is answered extensively due to its good None-linear approximation and generalization ability With nonlinear auto-companding sliding average (the Nonlinear Auto regressive Moving with feedback linearization Average with Feedback Linearization, NARMA-L2) controller is a kind of effective artificial neural network control Device framework processed.Under certain condition, the input/output relation of nonlinear system can be obtained by NARMA-L2 Model Distinguish, and Control law can be obtained by simple mathematic(al) manipulation.However since NARMA-L2 model is there are modeling error and training error, It is affected so that designed controller performance is applied in envelope curve.Therefore, present invention trial propose it is a kind of have repair online Controller (OC-NARMA-L2) is alleviated in the aero-engine performance decline of positive ability, and this method is repaired online using gradient descent method The neural network parameter of positive control device makes it have adaptive characteristic.

Summary of the invention

In view of the above technical problems, the present invention provides a kind of ANN Control side that aeroengine thrust decline is alleviated Method will be based on aiming at the problem that in the case where engine components performance is mutated, thrust level declines conventional multivariable controller The outer ring instruction correction module for becoming increment LP optimization algorithm is combined with inner loop control system, obtains the thrust with self-adjusting ability Nerve network controller is alleviated in decline.This method can be alleviated in the case where not overtemperature, not excess revolutions is dashed forward by engine components performance Complete machine thrust loss caused by becoming improves the performance of engine overall performance.

Technical solution: to achieve the above object, the technical solution adopted by the present invention are as follows:

A kind of neural network control method that aeroengine thrust decline is alleviated, which comprises the following steps:

Step A) establish physical descriptor interpolation mould between engine high pressure rotor speed, pressure ratio and jet pipe throat area Type, and using the relationship between main fuel amount and high pressure rotor revolving speed, using in the alleviation of neural network performance degradation Ring controls NARMA-L2 model；

Step B) NARMA-L2 model is combined, it designs and correction module is instructed based on the outer ring for becoming increment LP algorithm, exploitation exists Nerve network controller is alleviated in the thrust decline with self-adjusting ability in the case that engine generation part performance is mutated.

Step C) engine generation part performance mutation when, by outer ring be added instruction correction module, to engine instruct It is modified, alleviates the mesh that nerve network controller tracks revised instruction and then reaches thrust recovery using thrust decline 's.

Further, the step A) in using the relationship between main fuel amount and high pressure rotor revolving speed, using nerve net Network establishes the inner loop control NARMA-L2 model that performance degradation is alleviated, and specific step is as follows:

Step A1), choose engine main fuel amount as input quantity u [k], high pressure rotor revolving speed as output quantity y [k], And NARMA-L2 model is constructed according to selected input and output amount are as follows:

Y [k]=f₀(y[k-1],y[k-2],…,y[k-n],u[k-1],u[k-2],…,u[k-n])

+g₀(y[k-1],y[k-2],…,y[k-n],u[k-1],u[k-2],…,u[k-n])u[k]

Wherein, f₀And g₀Training neural network group f can be passed through₀ ^*And g₀ ^*Approximation obtains, and revolving speed control law are as follows:

Wherein, y^*[k] is engine desired output；And according to the instruction of engine complete machine blow down ratio, reality output blow down ratio It converts to obtain jet pipe throat area control amount with output revolving speed interpolation；

Step A2), utilize the error e between desired output and reality output_c(k) to neural network group f₀ ^*And g₀ ^*Topology knot Structure parameter carries out on-line amending:

W (k+1)=W (k)+α Δ W (k)

V (k+1)=V (k)+α Δ V (k)

Wherein, W and V is respectively the weight and threshold value of neural network group；α is learning rate, the adjustment of size affecting parameters Speed follows the example of determination by repeatedly trying；Δ W (k) and Δ V (k) is Jacobian matrix.Each element in Topology Vector is at the k moment It updates as follows:

w_i(k+1)=w_i(k)-αΔw_i(k)

v_i(k+1)=v_i(k)-αΔv_i(k)

Wherein, T={ y (k), y (k-1), y (k-2) ..., y (k-n), u (k-1), u (k-2) ..., u (k-n) }.

Further, the step B) it is middle in conjunction with NARMA-L2 model, it designs based on the outer ring instruction for becoming increment LP algorithm Correction module obtains the thrust decline in the case where engine generation part performance is mutated with self-adjusting ability and alleviates nerve Specific step is as follows for network controller:

Step B1), it constructs and nerve network control system structure is alleviated based on the aeroengine thrust for becoming increment LP optimization Figure, including the instruction amendment of NARMA-L2 model, inner ring nerve network controller, engine mockup, performance state filter, outer ring Module；

Step B2), engine initial order is determined according to flying condition and relevant control plan, utilizes engine mockup Output and restrictive condition design are established based on the outer ring instruction correction module for becoming increment LP algorithm in engine generation part performance Nerve network controller is alleviated in thrust decline in the case where mutation with self-adjusting ability.

Further, the step B2) it is based on becoming increment LP algorithm according to engine mockup output and restrictive condition design Outer ring instruct correction module, establish and decline in the case where the mutation of engine generation part performance with the thrust of self-adjusting ability Moving back alleviation nerve network controller, specific step is as follows:

Step B2.1), to not measurable propulsive effort and boundary restrictive condition, the small range near operating point is linearized, and is extracted Linear relationship PSM between performance indicator, amount of restraint and control amount, specific formula are as follows:

Wherein, Δ S_mfWith Δ S_mcThe respectively surge margin increment of fan and compressor, Δ n_LWith Δ n_HRespectively low pressure, High pressure rotor incremental speed, Δ F_nFor thrust increment, Δ T_t6For temperature increment after turbine, Δ W_fFor amount of fuel increment, Δ A₈For tail Area of injection orifice increment, P matrix element can be obtained from Stable status engine model by method of perturbation, specific formula are as follows:

P_ij={ M_i(u₀(j)+Δu_j,...)-M_i(u₀)}/Δu_j

I=1,2 ..., 6, j=1,2

Wherein, M_i() is the output matrix being calculated by model；u₀It is the initial value of control amount；Δu_jIt is j-th Control amount increment；

Step B2.2), for the control amount increment Delta u in step B2.1)_j, can be by corresponding control amount u_jMultiplied by increment coefficient It obtains, the selection rule of increment coefficient are as follows:

Wherein, k is the number of iterations；ΔF_nFor thrust increment.According to PSM (the propulsion system square extracted in step B2.1) Battle array, Propulsion System Matrix), instruction amendment is converted into LP optimization problem, solves instruction needed for thrust is alleviated Correction amount.

Further, the step B2.2) in instruction amendment is converted into LP optimization problem, solve needed for thrust is alleviated and refer to Specific step is as follows for the correction amount of order:

Step B2.2.1), in engine not excess revolutions, overtemperature, surge margin and control amount do not meet area requirement, thrust not More than searching engine maximum thrust, specific formula under the constraint of desired value are as follows:

Wherein, F_nFor motor power, W_fFor amount of fuel, A₈For nozzle area, u=[W_f,A₈]^TFor engine control to Amount, u_minAnd u_maxRespectively vector lower bound and the upper bound, n_LminAnd n_LmaxRespectively rotational speed of lower pressure turbine rotor is minimum and peak, n_Hmin And n_HmaxRespectively high pressure rotor revolving speed is minimum and peak, T_t6minAnd T_t6maxTemperature is minimum respectively after turbine and peak, S_mfAnd S_mcRespectively fan and compressor surge nargin, F_norThe engine for calculating the failure that do not mutate for real-time model pushes away Power desired value；

Step B2.2.2), according to control quantity constraint and restriction, using the PSM extracted in step B2.1), by step B2.2.1 the formula in) is rewritten are as follows:

Similarly acquire maximum thrust performance indicator are as follows:

ΔF_n=p₆₁ΔW_f+p₆₂ΔA₈

Step B2.2.3), solve LP problem, iterated to meet required precision or reach maximum number of iterations limitation After obtain globally optimal solution W_fAnd A₈, and optimal solution is input in engine, corresponding output signal n is calculated_HAnd EPR, Command signal makees poor, obtained Δ n before the two and thrust are restored_HrWith Δ EPR_rIt is instructed as rotary speed instruction and complete machine blow down ratio Correction value be input to inner ring nerve network controller, the instruction after inner ring nerve network controller tracking correction, alleviation is started Machine complete machine thrust loss.

Further, step C) in, when engine generation part performance is mutated, engine speed instruction is first declined 1%, Then outer ring is added and repairs instruction.

The utility model has the advantages that the invention proposes alleviate ANN Control based on the aeroengine thrust for becoming increment LP algorithm Device, design instruction amendment outer ring circuit on the basis of multivariable controller, wherein revolving speed control pass through neural network learning NARMA-L2 model inversion is realized.When engine performance mutates, first high pressure rotor rotary speed instruction reduction by 1% is ensured to send out Motivation trouble free service obtains high pressure rotor revolving speed under guarantee engine not out-of-limit condition using change increment LP optimization algorithm later The correction amount of instruction and the instruction of complete machine blow down ratio, so that engine be made to continue to push away needed for providing as far as possible in performance degradation Power.

Detailed description of the invention

Fig. 1 is that the present invention is based on the aeroengine thrusts for becoming increment LP optimization to alleviate nerve network control system structure Figure.

Fig. 2 is inner ring NARMA-L2 control system architecture figure.

Fig. 3 is to become increment LP algorithm structure figure.

Fig. 4 is H=0, and Ma=0, fuel delivery compares figure when performance mutation occurs for fan.

Fig. 5 is H=0, and Ma=0, high pressure rotor revolving speed compares figure when performance mutation occurs for fan.

Fig. 6 is H=0, and Ma=0, thrust compares figure when performance mutation occurs for fan.

Fig. 7 is H=8000m, and Ma=1.2, fuel delivery compares figure when performance mutation occurs for compressor.

Fig. 8 is H=8000m, and Ma=1.2, high pressure rotor revolving speed compares figure when performance mutation occurs for compressor.

Fig. 9 is H=8000m, and Ma=1.2, thrust compares figure when performance mutation occurs for compressor.

Specific embodiment

A specific embodiment of the invention is further described with reference to the accompanying drawing.

The neural network control method that a kind of aeroengine thrust decline that the present invention illustrates is alleviated, including following step It is rapid:

Step A) establish physical descriptor interpolation mould between engine high pressure rotor speed, pressure ratio and jet pipe throat area Type, and using the relationship between main fuel amount and high pressure rotor revolving speed, using in the alleviation of neural network performance degradation Ring controls NARMA-L2 model；

Step B) NARMA-L2 model is combined, it designs and correction module is instructed based on the outer ring for becoming increment LP algorithm, exploitation exists Nerve network controller is alleviated in the thrust decline with self-adjusting ability in the case that engine generation part performance is mutated.

Step A1), choose engine main fuel amount as input quantity u [k], high pressure rotor revolving speed as output quantity y [k], And NARMA-L2 model is constructed according to selected input and output amount are as follows:

Y [k]=f₀(y[k-1],y[k-2],…,y[k-n],u[k-1],u[k-2],…,u[k-n])+g₀(y[k-1],y [k-2],…,y[k-n],u[k-1],u[k-2],…,u[k-n])u[k]

Wherein, f₀And g₀Training neural network group f can be passed through₀ ^*And g₀ ^*Approximation obtains, and revolving speed control law are as follows:

Wherein, y^*[k] is engine desired output；And according to the instruction of engine complete machine blow down ratio, reality output blow down ratio It converts to obtain jet pipe throat area control amount with output revolving speed interpolation；

Step A2), utilize the error e between desired output and reality output_c(k) to neural network group f₀ ^*And g₀ ^*Topology knot Structure parameter carries out on-line amending:

W (k+1)=W (k)+α Δ W (k)

V (k+1)=V (k)+α Δ V (k)

Wherein, W and V is respectively the weight and threshold value of neural network group；α is learning rate, the adjustment of size affecting parameters Speed follows the example of determination by repeatedly trying；Δ W (k) and Δ V (k) is Jacobian matrix.Each element in Topology Vector is at the k moment It updates as follows:

w_i(k+1)=w_i(k)-αΔw_i(k)

v_i(k+1)=v_i(k)-αΔv_i(k)

Wherein, T={ y (k), y (k-1), y (k-2) ..., y (k-n), u (k-1), u (k-2) ..., u (k-n) }.

Step B1), it constructs and nerve network control system structure is alleviated based on the aeroengine thrust for becoming increment LP optimization Figure, including the instruction amendment of NARMA-L2 model, inner ring nerve network controller, engine mockup, performance state filter, outer ring Module；

Step B2), engine initial order is determined according to flying condition and relevant control plan, utilizes engine mockup Output and restrictive condition design are established based on the outer ring instruction correction module for becoming increment LP algorithm in engine generation part performance Nerve network controller is alleviated in thrust decline in the case where mutation with self-adjusting ability.

Step B2.1), to not measurable propulsive effort and boundary restrictive condition, the small range near operating point is linearized, and is extracted Linear relationship PSM between performance indicator, amount of restraint and control amount, specific formula are as follows:

Wherein, Δ S_mfWith Δ S_mcThe respectively surge margin increment of fan and compressor, Δ n_LWith Δ n_HRespectively low pressure, High pressure rotor incremental speed, Δ F_nFor thrust increment, Δ T_t6For temperature increment after turbine, Δ W_fFor amount of fuel increment, Δ A₈For tail Area of injection orifice increment, P matrix element can be obtained from Stable status engine model by method of perturbation, specific formula are as follows:

P_ij={ M_i(u₀(j)+Δu_j,...)-M_i(u₀)}/Δu_j

I=1,2 ..., 6, j=1,2

Wherein, M_i() is the output matrix being calculated by model；u₀It is the initial value of control amount；Δu_jIt is j-th Control amount increment；

Step B2.2), for the control amount increment Delta u in step B2.1)_j, can be by corresponding control amount u_jMultiplied by increment coefficient It obtains, the selection rule of increment coefficient are as follows:

Wherein, k is the number of iterations；ΔF_nFor thrust increment.According to the PSM extracted in step B2.1), instruction amendment is turned It is changed to LP optimization problem, solves the correction amount instructed needed for thrust is alleviated.

Step B2.2.1), in engine not excess revolutions, overtemperature, surge margin and control amount do not meet area requirement, thrust not More than searching engine maximum thrust, specific formula under the constraint of desired value are as follows:

Wherein, F_nFor motor power, W_fFor amount of fuel, A₈For nozzle area, u=[W_f,A₈]^TFor engine control to Amount, u_minAnd u_maxRespectively vector lower bound and the upper bound, n_LminAnd n_LmaxRespectively rotational speed of lower pressure turbine rotor is minimum and peak, n_Hmin And n_HmaxRespectively high pressure rotor revolving speed is minimum and peak, T_t6minAnd T_t6maxTemperature is minimum respectively after turbine and peak, S_mfAnd S_mcRespectively fan and compressor surge nargin, F_norThe engine for calculating the failure that do not mutate for real-time model pushes away Power desired value；

Step B2.2.2), according to control quantity constraint and restriction, using the PSM extracted in step B2.1), by step B2.2.1 the formula in) is rewritten are as follows:

Similarly acquire maximum thrust performance indicator are as follows:

ΔF_n=p₆₁ΔW_f+p₆₂ΔA₈

Step B2.2.3), solve LP problem, iterated to meet required precision or reach maximum number of iterations limitation After obtain globally optimal solution W_fAnd A₈, and optimal solution is input in engine, corresponding output signal n is calculated_HAnd EPR, Command signal makees poor, obtained Δ n before the two and thrust are restored_HrWith Δ EPR_rIt is instructed as rotary speed instruction and complete machine blow down ratio Correction value be input to inner ring nerve network controller, the instruction after inner ring nerve network controller tracking correction, alleviation is started Machine complete machine thrust loss.

In order to verify aeroengine thrust designed by the present invention decline alleviate neural network control method it is effective Property, the Digital Simulation that control is alleviated in thrust decline in certain envelope curve has been carried out under MATLAB environment.

The present invention is using the non-linear components model of certain small Bypass Ratio Turbofan Engine of type birotor as controlled device.It should Model is constructed by the method for Object-Oriented Programming, includes air intake duct, fan, compressor, combustion chamber, turbine and jet pipe etc. The important component of aero-engine, and be easy to call in MATLAB environment.

Before carrying out thrust and alleviating control, it is necessary first to which training obtains inner ring nerve network controller.Wherein, controller In two neural networks of NARMA-L2 model be single hidden layer, and the neuron of input layer, hidden layer and output layer Number is respectively 5,10 and 1.Training firstly the need of using aero-engine model at H=0, the flying condition of Ma=0, by with The input signal of machine generates main fuel amount-high pressure rotor rotary speed data collection comprising 10000 groups of training samples, wherein stochastic inputs Signal W_fRange are as follows: 96.79% to 100%, model output signal n_HRange are as follows: 99.42% to 100%.Training is completed Afterwards, in order to further obtain the controller with adaptive ability, on-line amending is carried out to neural network parameter, wherein learning rate α=0.0022, so far inner ring Design of Neural Network Controller finishes, and structure is as shown in Figure 2.

Outer ring instruction amendment is added on the basis of interior ring controller, obtains complete thrust decline and alleviates controller and needle Simulating, verifying is carried out to the controller, wherein outer ring instructs modified change increment LP algorithm structure as shown in Figure 3.Emulation shares 2 Group, comprising 2 flying conditions, respectively H=0, Ma=0 and H=8000m in envelope curve, Ma=1.2, wherein the 1st group of emulation is high Rotor speed is pressed to instruct n_HrInitial value is 1, and performance mutation occurs for engine blower after 10s, and efficiency reduces by 5%, after 20s Start revision directive, and emulates and terminate in 30s；2nd group of emulation n_HrInitial value is 0.95, after 10s engine blower and Performance mutation occurs for compressor, and the two flow and efficiency reduce by 1%, starts revision directive after 20s, equally imitates in 30s Really terminate.

Fig. 4-6 illustrates H=0, Ma=0, n_HrSelected variable change when performance mutation occurs for fan at=1 stable operating point Situation.3 lines in figure respectively correspond three kinds of situations: 1. in the case of engine work, conventional inner loop control side The variation of main fuel, high pressure rotor revolving speed and thrust under formula；In the case that 2. performance mutation occurs for engine, conventional inner ring control The variation of main fuel, high pressure rotor revolving speed and thrust under mode processed；3. in the case where engine mutation failure, thrust is alleviated The variation of main fuel, high pressure rotor revolving speed and thrust under control mode.

As can be seen from Figure 4 after performance mutation occurs, conventional rotating speed control guarantees high pressure rotor by reducing amount of fuel Revolving speed continues to track former instruction, this causes motor power to fail.Control method is alleviated in thrust decline proposed by the present invention First by n after mutation occurs_Hr1% is reduced to ensure that engine health works, and in 20s that revised instruction input is neural Network controller carries out thrust recovery, but from figure 5 it can be seen that due to high pressure rotor revolving speed maximum value n at this time_Hmax= 1.018, lead to the n in becoming increment LP optimization algorithm searching process_HReach boundary and optimizing stops.Though from Fig. 6 it can be found that Right thrust alleviates the thrust under control and thrust when not recovering to without mutation, but compared to conventional control, still is guaranteeing to send out Motivation makes thrust get a promotion in the case where not transfiniting.

Fig. 7-9 illustrates H=8000, Ma=1.2, n_HrPerformance mutation when institute occurs for compressor at=0.95 stable operating point Select variable change situation.With the promotion of height and Mach number, engine characteristics have been varied widely compared to ground point. Beneficial to the adaptive characteristic of inner ring nerve network controller, so that controller still can be very good tracking revolving speed after performance mutation Instruction, but at this time thrust size compared to engine work when declined.After outer ring instruction corrective loop is added, it is The work of guarantee engine health, rotary speed instruction signal first lower 1%, and at this time controller tracks new command signal quickly, Later using command signal increment needed for increment LP algorithm obtains thrust alleviation is become, in the Δ n that 20s obtains iteration_HrWith ΔEPR_rInner loop control device is acted on simultaneously, it can be seen in figure 9 that the thrust designed herein alleviates control method compared to normal Rule inner loop control device can effectively restore thrust.

Aiming at the problem that present invention thrust decline caused by aero-engine performance mutation causes component capabilities to be degraded, propose A kind of neural network control method alleviated based on the aeroengine thrust decline for becoming increment LP optimization.This method is according to hair Motivation high pressure rotor revolving speed and the instruction of complete machine blow down ratio, calculate fuel flow and nozzle by engine multivariable controller Throat area controls variable, and wherein inner loop control is realized by the NARMA-L2 model inversion of neural network learning；In multivariable control Design instruction amendment outer ring circuit on the basis of device processed, is mutated using change increment LP optimization method, and according to engine components performance Caused thrust recession level adjusts engine command signal to alleviate complete machine thrust loss.Simulation results show design Thrust under component capabilities sudden change conditions alleviates the validity of control method.

It should be pointed out that the above description is merely a specific embodiment, but protection scope of the present invention is not Be confined to this, anyone skilled in the art in the technical scope disclosed by the present invention, the change that can be readily occurred in Change and replace, should be covered by the scope of protection of the present invention.Therefore, protection scope of the present invention should be with the claim Subject to protection scope.