CN106483850A - The Fuzzy Self-adaptive PID method for designing that a kind of aero-engine is feedovered based on RBF neural - Google Patents

The Fuzzy Self-adaptive PID method for designing that a kind of aero-engine is feedovered based on RBF neural Download PDF

Info

Publication number
CN106483850A
CN106483850A CN201611065365.8A CN201611065365A CN106483850A CN 106483850 A CN106483850 A CN 106483850A CN 201611065365 A CN201611065365 A CN 201611065365A CN 106483850 A CN106483850 A CN 106483850A
Authority
CN
China
Prior art keywords
fuzzy
engine
network
aero
function
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN201611065365.8A
Other languages
Chinese (zh)
Inventor
申春艳
姜锐
景月
汤庚
韩冬
张景峰
张春艳
王惠
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shenyang Aerospace Xinguang Group Co Ltd
Original Assignee
Shenyang Aerospace Xinguang Group Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shenyang Aerospace Xinguang Group Co Ltd filed Critical Shenyang Aerospace Xinguang Group Co Ltd
Priority to CN201611065365.8A priority Critical patent/CN106483850A/en
Publication of CN106483850A publication Critical patent/CN106483850A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
    • G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
    • G05B13/04—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators
    • G05B13/042—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators in which a parameter or coefficient is automatically adjusted to optimise the performance

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Computation (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Automation & Control Theory (AREA)
  • Feedback Control In General (AREA)

Abstract

The invention discloses the Fuzzy PID self-adaptive control device method for designing that a kind of aero-engine is feedovered based on RBF neural, belong to Aeroengine Control Systems field, for improving aero-engine and its part acceleration and deceleration working condition in the presence of external condition and internal control, engine is made to have good acceleration and deceleration characteristic.Technical scheme includes：Based on aero-engine stable state firing test data sample, off-line training is carried out using RBF neural, constitute feedforward controller, add Fuzzy Self-adaptive PID on this basis.The present invention is by carrying out feed speed control to each state of aero-engine, controller has preferable tracking effect, controller parameter on-line tuning ability, system be can further improve in large-scale Control platform, the control that the nonlinear characteristic of effectively solving engine is brought is difficult.

Description

The Fuzzy Adaptive PID Control that a kind of aero-engine is feedovered based on RBF neural Device method for designing
Technical field
The invention discloses the Fuzzy Self-adaptive PID that a kind of aero-engine is feedovered based on RBF neural sets Meter method, belongs to Aeroengine Control Systems field, for improving aero-engine and its part in external condition and inside Acceleration and deceleration working condition in the presence of control, makes engine have good acceleration and deceleration characteristic.
Background technology
Aero-engine and its part have the working range of broadness in the presence of external condition and internal control.Outward Boundary's condition mainly has：The engine intake stagnation temperature determined by flight M number and flying height and stagnation pressure etc..Its internal control effect master Have：Main chamber and after-burner amount of fuel, the throat area of reaction jet pipe, blade (fan, compressor, whirlpool Wheel) established angle etc..The collective effect of external condition and internal control factor causes engine characteristic in the course of the work to become Change amplitude is very big, and the process is the complicated aerothermodynamics course of work.This broad working range necessarily causes its tool There is strong nonlinear characteristic.Therefore, if not implementing control to such process, the normal work of aero-engine cannot just be ensured Make.In order that each part of engine all can be stablized under any environmental condition and any working condition, reliably run, must Rationally control must be implemented to which, so as to its benefit in performance can be given full play to.
As aero-engine constantly develops towards high-performance, high reliability, efficient direction, aero-engine controls Systematic research is also constantly deeply.And the controller design method of main flow is that nonlinear system is converted into linear system at present System, so as to using based on lineary system theory method come be controlled device design.And be one in aero-engine system nature The system with nonlinearity is planted, system performance parameter has very big change with the change of environmental condition and working condition Change scope.If not considering its nonlinear characteristic in controller design, will be unable to ensure control effect.At present, Linear Control Method has formed the theory of complete set, and the research for being controlled system design to linear system is quite ripe, and for non- The design of linear controller does not also form the theoretical system of complete set.Non-linear control is designed with general linear control theory Device processed can not meet demand of the system to control performance on a large scale well, it would be highly desirable to need using new Design of non-linear controllers Method is ensureing the performance of system.
In recent years, with the continuous development of aero-engine Full Authority Digital Electronic Control research, associated control The research of algorithm is also more and more deep.As conventional PID controllers simple structure is easily achieved, and robustness is good, so PID control Device processed is still used widely in practice in current engineering.But aero-engine has strong nonlinear characteristic, control The conventional PID controllers that parameter is fixed cannot meet its control performance and require.Fixing PID controller parameter is only applicable to necessarily Working range in or set point, and obtain controller parameter when need substantial amounts of manpower and time.Aero-engine is whole When flight envelope works, as engine nonlinear characteristic affects, when dynamic characteristic is unknowable or occurs uncertain During change, linear controller is difficult to obtain good control effect, and it cannot be guaranteed that the reliability of control system, causes control product Matter is not good enough.For these reasons, so beginning with nonlinear control techniques, such as F100, the F110 of active service operational aircraft, In the engines such as F404, have started to attempt application nonlinear control algorithm, NASA is mounted with the numeral of F100 on F-15 aircraft Electronic controller (DEEC), has carried out the research of Dynamic matrix control pattern, including：Integrated flight Solid rocket engine, adaptive engine Control, engine lengthens the life control and air intake duct Comprehensive Control etc., and by Flight the performance of controller [11].Although In this way, domestic and international research or very limited amount of, answering mainly due to Non-Linear Control Theory to nonlinear control techniques Polygamy, it are difficult to be applied in control systems engineering (CSE).But with the research of aero-engine Full Authority Digital Electronic Control not Disconnected development, many scholars both domestic and external begin to focus on the method for designing of Non-Linear Control Theory.The control algolithm of main flow is big at present In terms of all concentrating on the nonlinear control algorithms such as Self Adaptive Control and ANN Control.
The principle of fuzzy control be by by related to Fuzzy Set Theory and fuzzy logic inference etc. for control method reason By combining, the fuzzy thinking to people is simulated, so as to solve the control of some objects that cannot be described by Mathematical Modeling System design problem processed.Fuzzy control belongs to nonlinear control method, and this method has working range width and do not rely on The advantages of Mathematical Modeling, it is widely used in engineering practice.In design of Fuzzy Control, not yet set up at present Playing effective method carries out rule settings, no matter is required for dependence experience and examination to gather using control table or control analytic formula, There is certain limitation.In order to overcome this shortcoming of fuzzy control, often the control is combined with other control forms, wherein The most common are fuzzy controller.
Different according to structure is constituted, fuzzy-adaptation PID control can be classified as two main types.One class is PID type fuzzy control Fuzzy domination structure is constituted the structure with PID function for device, this quasi-controller.Its input is system deviation and change of error Rate, controller are output as controlled quentity controlled variable or controlled quentity controlled variable rate of change, and it is by the fuzzy reasoning process based on expertise, makes control Device processed is output as nonlinear function, and this quasi-controller is referred to as fuzzy controller, because the controller is control structure being similar to In PID controller structure, and actually do not have proportionality coefficient, integral coefficient and the differential coefficient of conventional PID controller, take and Instead of be a series of based on expert adjust Heuristics constitute fuzzy inference rule.Another kind of for Fuzzy Adaptive PID control Device processed.The controller is the controller being bonded by conventional PID controller and fuzzy reasoning control.Base with fuzzy control This theory and method, condition rule, operation with obscuring set representations, and using these fuzzy control rules and for information about as Knowledge is stored in computer literacy storehouse.According to the fuzzy relation between systematic error and error rate, obtained by fuzzy reasoning To proportionality coefficient, three Parameters variation amounts of integral coefficient and differential coefficient, to meet different errors and error rate to three The different requirements of parameter, make control system have good dynamic and static properties.
ANN Control is that a kind of have the advanced of self adaptation and self-learning capability to complicated uncertain problem Control theory, this control method can be approached to Any Nonlinear Function with arbitrary precision, had simultaneously for information Very strong comprehensive treatment capability, can realize while coordinating to the relation between a variety of input informations and processing.At present Neutral net has been applied in the control of complex object, the non-linear and self-organizing of its essence, self-learning capability.
The present invention utilizes the Fuzzy Self-adaptive PID method for designing feedovered based on RBF neural, for improving boat Empty engine and its part acceleration and deceleration working condition in the presence of external condition and internal control, has engine good Acceleration and deceleration characteristic.The controller has more preferable tracking effect, controller parameter on-line tuning ability, can further improve system In large-scale Control platform, the control difficulty that can be brought with the nonlinear characteristic of effectively solving engine.
Content of the invention
The technical problem to be solved allows for aero-engine and its part in external condition and internal control There is in the presence of system the working range of broadness so that characteristic variations amplitude of the engine in this working range is very big, and This process is the complicated aerothermodynamics course of work, with strong nonlinear characteristic, designs and is feedovered based on RBF neural Fuzzy Self-adaptive PID method for designing, for improving aero-engine and its part in external condition and internal control In the presence of acceleration and deceleration working condition, make engine have good acceleration and deceleration characteristic.Controller is made to have preferably tracking effect Really, controller parameter on-line tuning ability, can further improve system in large-scale Control platform, can be started with effectively solving The control that the nonlinear characteristic of machine is brought is difficult.
, it is ensured that aero-engine is stable, reliably work.
The present invention is adopted the following technical scheme that：The fuzzy that a kind of aero-engine is feedovered based on RBF neural is adaptive Controller design is answered, is comprised the steps
Step 1), alternative pack level model is used as aero-engine nonlinear model；
Step 2), for step 1) in nonlinear model design aero-engine Fuzzy Self-adaptive PID, utilize Fuzzy control law makes controller parameter on-line tuning；
Step 3), based on aero-engine total state test run steady state data, reflected using powerful non-linear of RBF neural Ability is penetrated, trains the feedforward controller of exportable accurate feedforward amount；
Step 4), RBF neural feedforward control system is combined with Fuzzy Adaptive PID feedback control system, in boat Emulated on empty engine nonlinear model.
Further, the step 2) in nonlinear model design aero-engine Fuzzy Self-adaptive PID design Step is as follows：
Using most basic control law, i.e.,
U (t)=kpe(t)+ki∫e(t)dt+kdde(t)/dt
Wherein kp、ki、kdFor three control parameters.Fuzzy inference system is with error e and error rate evAs input, Using fuzzy reasoning method to pid parameter kp、ki、kdOn-line tuning is carried out, to meet in different error e and error rate evIn the case of to controller parameter difference requirement
The fuzzy controller of pid parameter adjustment is in the form of two inputs, two output.The controller be with error e and error Rate of change evAs input, the correction Δ k of two parameters P, I of PID controllerp、ΔkiAs output.Take error originated from input e and mistake Difference rate of change evAnd output Δ kp、ΔkiFuzzy subset is { NB, NM, NS, ZO, PS, PM, PB }, and in subset, element represents negative respectively Greatly, in bearing, negative little, zero, just little, center, honest.Error e and error rate evDomain be [- 3,3], quantification gradation for- 3,-2,-1,0,1,2,3}.
According to degree of membership assignment table and each parameter fuzzy Controlling model of each fuzzy subset, fuzzy reasoning is applied to calculate Pid parameter variable quantity, substitutes into following equation and calculates：
kp=kp0+Δkp
ki=ki0+Δki
In formula：kp0、ki0For the initial design values of pid parameter, designed by the parameter tuning method of conventional PID controller. Δkp、ΔkiExport for 2 of fuzzy controller, can taking according to two control parameters of state adjust automatically PID of controlled device Value, takes kp0Value be 0.00006, ki0Value be taken as 0.0005.
Error, the domain of error rate and output Δ kp、ΔkiAfter fuzzy subset determines, the person in servitude of fuzzy variable need to be determined Membership fuction.I.e. to fuzzy variable assignment, degree of membership of the domain interior element to fuzzy variable is determined.Person in servitude in fuzzy logic toolbox In category degree Functions editor, input quantity e, e is selectedvMembership function be Gaussian (gaussmf) and triangle (trimf), defeated Go out Δ kp、ΔkiMembership function be Gaussian (gaussmf) and triangle (trimf).
Further, the step 3) aero-engine total state test run steady state data is based on, powerful using neutral net Non-linear mapping capability, train the Feedforward Controller Design step of exportable accurate feedforward amount as follows：
From RBF neural, this network is three layers of feedforward network with single hidden layer, not only can be forced with arbitrary accuracy Nearly arbitrary function, and also there is in terms of pattern classification certain advantage.Three-layer network is respectively input layer, hidden layer, defeated Go out layer.Signal passes to hidden layer by input layer；Determine hidden layer node property is to imply layer functions, chooses radially Basic function；Linear operation is completed by output layer from the signal of hidden layer output.
Part I is hidden layer, can realize the Nonlinear Mapping from input layer to output layer.From basic function it is：
In formula：The center vector of j-th node of network is Cj=[cj1,cj2,...,cji,...,cjn]T, wherein i=1, 2,...,N；The sound stage width vector of network is B=[b1,b2,...,bj,...,bp]T, bjVectorial for the sound stage width of node j, its value is permissible Unrestricted choice, and it is more than zero, being determined width of the basic function around center by it, the width show also radial direction base neuron Susceptibility, for given input X, only sub-fraction center is activated near the processing unit of X, completes partial approximation；| |X-Cj| | for vectorial X-CjNorm, represent X and CjBetween distance；fjIt is radial symmetric function (Gaussian function), the letter Number is in CjThere is a unique maximum at place, and with | | X-Cj| | increase, fjZero is decayed to rapidly.
Part II is output layer：Realize hj(x)→y1Mapping, expression formula is as follows：
y1(k)=w1h1+w2h2+...+wphp
In formula：W=[w1,w2,...,wp]TOutput weight vector for network.
Gaussian function constructing RBF neural, i.e.,
From above formula, Gaussian bases form is simple, even if also will not be too complicated when input variable is more.In addition may be used Know that RBF is the nonlinear balance attenuation function of non-negative of radial symmetric, and as its line smoothing is good, so Any order derivative of the RBF is all present.
In RBF neural application, the parameter of neutral net is affected to have：Hidden neuron interstitial content, basic function Center vector CjAnd the weight w between connection hidden layer and output layer etc..For the number of hidden layer neuron, it can be made For a learning parameter, it is also possible to chosen by experience, neuron number is more, then the operational capability of neutral net is stronger, That is the approximation capability of RBF is stronger.
A radial primary function network is quickly designed from newrbe function.When radial basis function network is built, newrbe letter Number can increase neuron number automatically according to input vector, farthest reduce error so that design error is close to 0. Continue if the not up to required precision to increase neuron, meet then network design success after required precision.Program determination condition It is to meet required precision or reach maximum neuron number.When application newrbe function carries out radial primary function network design, Distribution density spread is a very important parameter, and when network training is carried out, suitable distribution density will affect network to survey Examination precision.
The input layer number of RBF neural is depending on the number of input vector.The present invention is from aero-engine Under the above state of slow train, from 50 (kr/min) to 116, (kr/min) takes a sample point every 1 (kr/min), and sample point is current Stable state input fuel flow Wf and steady-state speed n under state.Sample data includes that (stable state is input into fuel flow to object vector T Wf) and input sample P (steady-state speed n), sample size be 66.As distribution density spread of RBF cannot be prior Determine and spread affect network precision, therefore network foundation during set it to 1,2,3,4 and 5, totally 5 whole Number, observes their impacts to neural network forecast performance, and network naming is net.
Description of the drawings
Fig. 1 is the controller architecture figure of the present invention.
Fig. 2 is the Fuzzy Adaptive PID workflow diagram of the present invention.
Fig. 3 is the Stepped Impedance Resonators simulation curve figure of the present invention.
Phantom error figure when Fig. 4 is the Stepped Impedance Resonators of the present invention.
Δ K when Fig. 5 a, 5b are the Stepped Impedance Resonators of the present invention respectivelyp、ΔKiChange curve.
Fig. 6 is the ladder input simulation curve figure of the present invention.
Fig. 7 is the phantom error figure when ladder of the present invention is input into.
Fig. 8 a, 8b are Δ K when the ladder of the present invention is input intop、ΔKiChange curve.
Specific embodiment
Specific embodiment to the present invention elaborates below in conjunction with the accompanying drawings, so as to do to protection scope of the present invention Go out apparent clearly to define.
The Fuzzy PID self-adaptive control device method for designing that a kind of aero-engine of the present invention is feedovered based on RBF neural, Comprise the steps：
Step 1), alternative pack level model is used as aero-engine nonlinear model；
Step 2), for step 1) in nonlinear model design aero-engine Fuzzy Self-adaptive PID, utilize Mould
Paste control law makes controller parameter on-line tuning；
Step 3), based on aero-engine total state test run steady state data, reflected using powerful non-linear of RBF neural Penetrate energy
Power, trains the feedforward controller of exportable accurate feedforward amount；
Step 4), RBF neural feedforward control system is combined with Fuzzy Adaptive PID feedback control system, in boat Emulated on empty engine nonlinear model.
Wherein step 2) in nonlinear model design aero-engine Fuzzy Self-adaptive PID method for designing as follows：
Step 2.1) adopt most basic control law, i.e.,
U (t)=kpe(t)+ki∫e(t)dt+kdde(t)/dt
Wherein kp、ki、kdFor three control parameters.Fuzzy inference system is with error e and error rate evAs input, Using fuzzy reasoning method to pid parameter kp、ki、kdOn-line tuning is carried out, to meet in different error e and error rate evIn the case of to controller parameter difference requirement.
Step 2.2) pid parameter adjustment fuzzy controller using two input two export in the form of.The controller be to miss Difference e and error rate evAs input, the correction Δ k of two parameters P, I of PID controllerp、ΔkiAs output.Take input Error e and error rate evAnd output Δ kp、ΔkiFuzzy subset is { NB, NM, NS, ZO, PS, PM, PB }, element in subset Represent respectively and bear greatly, in bearing, negative little, zero, just little, center, honest.Error e and error rate evDomain be [- 3,3], amount It is { -3, -2, -1,0,1,2,3 } to change grade.
Step 2.3) according to degree of membership assignment table and each parameter fuzzy Controlling model of each fuzzy subset, apply fuzzy reasoning Pid parameter variable quantity is calculated, is substituted into following equation and calculates：
kp=kp0+Δkp
ki=ki0+Δki
In formula：kp0、ki0For the initial design values of pid parameter, designed by the parameter tuning method of conventional PID controller. Δkp、ΔkiExport for 2 of fuzzy controller, can taking according to two control parameters of state adjust automatically PID of controlled device Value, takes kp0Value be 0.00006, ki0Value be taken as 0.0005.
Step 2.4) error, the domain of error rate and output Δ kp、ΔkiAfter fuzzy subset determines, need to determine fuzzy The membership function of variable.I.e. to fuzzy variable assignment, degree of membership of the domain interior element to fuzzy variable is determined.In fuzzy logic work In the membership function editing machine of tool case, input quantity e, e is selectedvMembership function be Gaussian (gaussmf) and triangle (trimf), Δ k is exportedp、ΔkiMembership function be Gaussian (gaussmf) and triangle (trimf).
Wherein step 3) aero-engine total state test run steady state data is based on, reflected using powerful non-linear of neutral net Ability is penetrated, trains the Feedforward Controller Design method of exportable accurate feedforward amount as follows：
Step 3.1) RBF neural is selected, this network is three layers of feedforward network with single hidden layer, not only can be to appoint Meaning precision approaches arbitrary function, and also has certain advantage in terms of pattern classification.Three-layer network is respectively input layer, hidden Containing layer, output layer.Signal passes to hidden layer by input layer；Determine hidden layer node property is to imply layer functions, Choose RBF；Linear operation is completed by output layer from the signal of hidden layer output.
Step 3.2) Part I is hidden layer, can realize the Nonlinear Mapping from input layer to output layer.From base Function is：
In formula：The center vector of j-th node of network is Cj=[cj1,cj2,...,cji,...,cjn]T, wherein i=1, 2,...,N；The sound stage width vector of network is B=[b1,b2,...,bj,...,bp]T, bjVectorial for the sound stage width of node j, its value is permissible Unrestricted choice, and it is more than zero, being determined width of the basic function around center by it, the width show also radial direction base neuron Susceptibility, for given input X, only sub-fraction center is activated near the processing unit of X, completes partial approximation；| |X-Cj| | for vectorial X-CjNorm, represent X and CjBetween distance；fjIt is radial symmetric function (Gaussian function), the letter Number is in CjThere is a unique maximum at place, and with | | X-Cj| | increase, fjZero is decayed to rapidly.
Step 3.3) Part II is output layer：Realize hj(x)→y1Mapping, expression formula is as follows：
y1(k)=w1h1+w2h2+...+wphp
In formula：W=[w1,w2,...,wp]TOutput weight vector for network.
Gaussian function constructing RBF neural, i.e.,
From above formula, Gaussian bases form is simple, even if also will not be too complicated when input variable is more.In addition may be used Know that RBF is the nonlinear balance attenuation function of non-negative of radial symmetric, and as its line smoothing is good, so Any order derivative of the RBF is all present.
In RBF neural application, the parameter of neutral net is affected to have：Hidden neuron interstitial content, basic function Center vector CjAnd the weight w between connection hidden layer and output layer etc..For the number of hidden layer neuron, it can be made For a learning parameter, it is also possible to chosen by experience, neuron number is more, then the operational capability of neutral net is stronger, That is the approximation capability of RBF is stronger.
Step 3.4) radial primary function network is quickly designed from newrbe function.When radial basis function network is built, Newrbe function can increase neuron number automatically according to input vector, farthest reduce error so that design error Close to 0.Continue if the not up to required precision to increase neuron, meet then network design success after required precision.Program End condition is to meet required precision or reach maximum neuron number.Application newrbe function carries out radial primary function network During design, distribution density spread is a very important parameter, and when network training is carried out, suitable distribution density is by shadow Ring network test precision.It is RBF neural establishment, training process below.
Step 3.5) RBF neural input layer number depending on input vector number.The present invention is from boat Sky starts under the above state of bicycle and motorcycle that (kr/min) takes a sample point, sample every 1 (kr/min) from 50 (kr/min) to 116 Point is input into fuel flow Wf and steady-state speed n for the stable state under current state.Sample data includes that (stable state is input into object vector T Fuel flow Wf) and input sample P (steady-state speed n), sample size be 66.Distribution density spread due to RBF Cannot be determined in advance and spread affect network precision, therefore network foundation during set it to 1,2,3,4 and 5, totally 5 integers, observe their impacts to neural network forecast performance, and network naming is net.
Step 3.6) Fuzzy Self-adaptive PID is added RBF feedforward controller, the controller is applied to engine On nonlinear model, input signal selects stairstep signal, big step signal respectively.
The Fuzzy PID self-adaptive control device method for designing that the present invention is feedovered based on RBF neural, first to aeroplane engine Machine nonlinear model carries out Fuzzy Self-adaptive PID design.This fuzzy inference system is with error e and error rate evMake For being input into, using fuzzy reasoning method to pid parameter kp、kiOn-line tuning is carried out, is become in different error e and error with meeting Rate evIn the case of difference requirement to controller parameter, and make controlled device have good dynamic, static properties.Utilize RBF neural carries out aero-engine inputoutput data fitting, improves the difficulty brought by fixing interpolation, so that gained is refreshing Through network at utmost approaching to reality system, train feedforward and measure and feedforward control system is built, anti-in conjunction with Fuzzy Adaptive PID Feedback control system carries out aero-engine control, makes system keep system stability in rotating speed wide variation, and with good Good dynamic property, and do not limited by operating point, so that parameter adaptive ability is improved further.
The above is only the preferred embodiment of the present invention, it is noted that for the ordinary skill people of the art For member, some improvement can also be made under the premise without departing from the principles of the invention, these improvement also should be regarded as the present invention's Protection domain.

Claims (3)

1. the Fuzzy Self-adaptive PID design that a kind of aero-engine is feedovered based on RBF neural, it is characterised in that Comprise the steps
Step 1), alternative pack level model is used as aero-engine nonlinear model；
Step 2), for step 1) in nonlinear model design aero-engine Fuzzy Self-adaptive PID, using fuzzy Control law makes controller parameter on-line tuning；
Step 3), based on aero-engine total state test run steady state data, using the powerful Nonlinear Mapping energy of RBF neural Power, trains the feedforward controller of exportable accurate feedforward amount；
Step 4), RBF neural feedforward control system is combined with Fuzzy Adaptive PID feedback control system, is sent out in aviation Emulated on motivation nonlinear model.
2. the Fuzzy Self-adaptive PID design that aero-engine as claimed in claim 1 is feedovered based on RBF neural Method, it is characterised in that：The step 2) in nonlinear model design aero-engine Fuzzy Self-adaptive PID design side Method is as follows：
Using most basic control law, i.e.,
U (t)=kpe(t)+ki∫e(t)dt+kdde(t)/dt
Wherein kp、ki、kdFor three control parameters.Fuzzy inference system is with error e and error rate evAs input, using mould Paste inference method is to pid parameter kp、ki、kdOn-line tuning is carried out, to meet in different error e and error rate evFeelings Difference requirement under condition to controller parameter,
The fuzzy controller of pid parameter adjustment is in the form of two inputs, two output.The controller be with error e and error change Rate evAs input, the correction Δ k of two parameters P, I of PID controllerp、ΔkiAs output.Take error originated from input e and error becomes Rate evAnd output Δ kp、ΔkiFuzzy subset is { NB, NM, NS, ZO, PS, PM, PB }, and in subset, element is represented respectively and born greatly, In negative, negative little, zero, just little, center, honest.Error e and error rate evDomain be [- 3,3], quantification gradation for -3, - 2,-1,0,1,2,3},
According to degree of membership assignment table and each parameter fuzzy Controlling model of each fuzzy subset, fuzzy reasoning is applied to calculate PID ginseng Number variable quantity, substitutes into following equation and calculates：
kp=kp0+Δkp
ki=ki0+Δki
In formula：kp0、ki0For the initial design values of pid parameter, designed by the parameter tuning method of conventional PID controller, Δ kp、 ΔkiExport for 2 of fuzzy controller, can according to the value of two control parameters of state adjust automatically PID of controlled device, Take kp0Value be 0.00006, ki0Value be taken as 0.0005,
Error, the domain of error rate and output Δ kp、ΔkiAfter fuzzy subset determines, need to determine fuzzy variable is subordinate to letter Number.I.e. to fuzzy variable assignment, degree of membership of the domain interior element to fuzzy variable is determined, in the degree of membership of fuzzy logic toolbox In Functions editor, input quantity e, e is selectedvMembership function be Gaussian (gaussmf) and triangle (trimf), export Δ kp、ΔkiMembership function be Gaussian (gaussmf) and triangle (trimf).
3. the Fuzzy Self-adaptive PID design that aero-engine as claimed in claim 1 is feedovered based on RBF neural Method, it is characterised in that：The step 3) aero-engine total state test run steady state data is based on, powerful using neutral net Non-linear mapping capability, trains the Feedforward Controller Design method of exportable accurate feedforward amount as follows：
From RBF neural, this network is three layers of feedforward network with single hidden layer, not only can be approached with arbitrary accuracy and appoint Meaning function, and also there is in terms of pattern classification certain advantage.Three-layer network is respectively input layer, hidden layer, output layer. Signal passes to hidden layer by input layer；Determine hidden layer node property is to imply layer functions, chooses radial direction base letter Number；Linear operation is completed by output layer from the signal of hidden layer output,
Part I is hidden layer, can realize the Nonlinear Mapping from input layer to output layer.From basic function it is：
In formula：The center vector of j-th node of network is Cj=[cj1,cj2,...,cji,...,cjn]T, wherein i=1,2 ..., N；The sound stage width vector of network is B=[b1,b2,...,bj,...,bp]T, bjVectorial for the sound stage width of node j, its value freely can be selected Select, and be more than zero, determined width of the basic function around center by it, the width show also the sensitivity of radial direction base neuron Degree, for given input X, only sub-fraction center is activated near the processing unit of X, completes partial approximation；||X-Cj| | for vectorial X-CjNorm, represent X and CjBetween distance；fjIt is radial symmetric function (Gaussian function), the function is in Cj There is a unique maximum at place, and with | | X-Cj| | increase, fjZero is decayed to rapidly,
Part II is output layer：Realize hj(x)→y1Mapping, expression formula is as follows：
y1(k)=w1h1+w2h2+...+wphp
In formula：W=[w1,w2,...,wp]TFor the output weight vector of network,
Gaussian function constructing RBF neural, i.e.,
From above formula, Gaussian bases form is simple, even if also will not be too complicated when input variable is more, in addition we know footpath It is the nonlinear balance attenuation function of non-negative of radial symmetric to basic function, and as its line smoothing is good, so the footpath All exist to any order derivative of basic function,
In RBF neural application, the parameter of neutral net is affected to have：Hidden neuron interstitial content, the center of basic function Vectorial CjAnd the weight w between connection hidden layer and output layer etc..For the number of hidden layer neuron, can be using it as one Individual learning parameter, it is also possible to chosen by experience, neuron number is more, then the operational capability of neutral net is stronger, also It is to say that the approximation capability of RBF is stronger,
A radial primary function network is quickly designed from newrbe function.When radial basis function network is built, newrbe function can Neuron number is increased automatically according to input vector, farthest reduce error so that design error close to 0, if Not up to required precision then continues to increase neuron, meets then network design success after required precision, and program determination condition is full Sufficient required precision reaches maximum neuron number, when application newrbe function carries out radial primary function network design, distribution Density spread is a very important parameter, and when network training is carried out, suitable distribution density will affect network test essence Degree,
The input layer number of RBF neural depends on the number of input vector, from the above shape of aeroplane engine bicycle and motorcycle Under state, from 50 (kr/min) to 116, (kr/min) takes a sample point every 1 (kr/min), and sample point is steady under current state State is input into fuel flow Wf and steady-state speed n.Sample data includes object vector T (stable state is input into fuel flow Wf) and input sample This P (steady-state speed n), sample size are 66, cannot be determined in advance due to distribution density spread of RBF and The precision of spread impact network, therefore sets it to 1,2,3,4 and 5, totally 5 integers, observation during network foundation Their impacts to neural network forecast performance, network naming are net.
CN201611065365.8A 2016-11-23 2016-11-23 The Fuzzy Self-adaptive PID method for designing that a kind of aero-engine is feedovered based on RBF neural Pending CN106483850A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201611065365.8A CN106483850A (en) 2016-11-23 2016-11-23 The Fuzzy Self-adaptive PID method for designing that a kind of aero-engine is feedovered based on RBF neural

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201611065365.8A CN106483850A (en) 2016-11-23 2016-11-23 The Fuzzy Self-adaptive PID method for designing that a kind of aero-engine is feedovered based on RBF neural

Publications (1)

Publication Number Publication Date
CN106483850A true CN106483850A (en) 2017-03-08

Family

ID=58275350

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201611065365.8A Pending CN106483850A (en) 2016-11-23 2016-11-23 The Fuzzy Self-adaptive PID method for designing that a kind of aero-engine is feedovered based on RBF neural

Country Status (1)

Country Link
CN (1) CN106483850A (en)

Cited By (22)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107203135A (en) * 2017-06-08 2017-09-26 东莞中子科学中心 A kind of adaptive control system feedovered based on dynamical feedback
CN107608215A (en) * 2017-10-18 2018-01-19 齐鲁工业大学 A kind of mapping UAV Attitude control method based on RBF neural
CN108828952A (en) * 2018-08-02 2018-11-16 中国民航大学 Steering engine Electric Loading System intelligent control method based on CMAC Neural Network
CN109308004A (en) * 2018-09-28 2019-02-05 东北大学 Loading procedure servo-control system and method behind a kind of true triaxial rigidity matching peak
WO2019144337A1 (en) * 2018-01-25 2019-08-01 大连理工大学 Deep-learning algorithm-based self-adaptive correction method for full-envelope model of aero-engine
CN110852421A (en) * 2019-11-11 2020-02-28 北京百度网讯科技有限公司 Model generation method and device
CN111046568A (en) * 2019-12-18 2020-04-21 哈尔滨工程大学 Control parameter optimization method based on aeroengine and jet nozzle electromechanical liquid multi-system combined simulation
CN111273544A (en) * 2020-04-01 2020-06-12 河海大学常州校区 Radar pitching motion control method based on prediction RBF feedforward compensation type fuzzy PID
CN111624886A (en) * 2020-06-05 2020-09-04 沈阳航空航天大学 Variable-cycle aero-engine thrust control method based on SARSA
CN111666648A (en) * 2019-12-04 2020-09-15 江西洪都航空工业集团有限责任公司 Method for simulating dynamic characteristics of aircraft engine
CN111734533A (en) * 2020-05-28 2020-10-02 北京仿真中心 Turbofan engine-based model prediction method and system
CN111738563A (en) * 2020-05-29 2020-10-02 中国航天标准化研究所 Method for detecting control capability of aerospace production field process
CN111880552A (en) * 2020-07-28 2020-11-03 中国科学院长春光学精密机械与物理研究所 Multi-rotor unmanned aerial vehicle trajectory tracking composite control method
CN113485443A (en) * 2021-08-10 2021-10-08 北京宇系航通科技有限公司 Unmanned aerial vehicle control method based on deep learning, storage medium and equipment
CN113759699A (en) * 2021-08-27 2021-12-07 北京工业大学 Servo motor rotating speed control method for reducing steady-state error
CN114687870A (en) * 2022-03-03 2022-07-01 沈阳航天新光集团有限公司 Vehicle-mounted power take-off generator set control system and control method
CN114706295A (en) * 2022-06-07 2022-07-05 中国飞机强度研究所 Method for controlling fast time-varying thermal load in thermal test for testing strength of aerospace plane
CN114815584A (en) * 2022-04-11 2022-07-29 哈尔滨工程大学 Circulating air injection amount feedforward PID closed-loop control method and system taking natural gas ejector inlet pressure fluctuation as input
CN114839860A (en) * 2022-04-11 2022-08-02 哈尔滨工程大学 Fuzzy PID fuel injection quantity closed-loop control method based on pressure change monitoring of high-pressure natural gas injector inlet
CN115236975A (en) * 2022-07-28 2022-10-25 南京工业大学 Missile pneumatic heat ground finite element analysis and IPD nonlinear sliding mode control method
WO2022252206A1 (en) * 2021-06-04 2022-12-08 大连理工大学 Aero-engine surge active control system based on fuzzy switching of controllers
CN117075481A (en) * 2023-10-13 2023-11-17 中国空气动力研究与发展中心高速空气动力研究所 Mach number composite control method for wind tunnel transonic ladder variable attack angle test

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102033548A (en) * 2009-09-29 2011-04-27 北京航空航天大学 RBF neural network-based servo control system and method
CN102331717A (en) * 2011-10-10 2012-01-25 哈尔滨工程大学 Intelligent control method of navigational speed of ship
CN105262396A (en) * 2015-11-26 2016-01-20 上海电机学院 PMSM (Permanent Magnet Synchronous Motor) control method based on HCMAC neural network
CN105629735A (en) * 2016-02-22 2016-06-01 南京航空航天大学 Online adaptive optimal PID controller design method for aeroengine

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102033548A (en) * 2009-09-29 2011-04-27 北京航空航天大学 RBF neural network-based servo control system and method
CN102331717A (en) * 2011-10-10 2012-01-25 哈尔滨工程大学 Intelligent control method of navigational speed of ship
CN105262396A (en) * 2015-11-26 2016-01-20 上海电机学院 PMSM (Permanent Magnet Synchronous Motor) control method based on HCMAC neural network
CN105629735A (en) * 2016-02-22 2016-06-01 南京航空航天大学 Online adaptive optimal PID controller design method for aeroengine

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
申春艳: "微型涡喷发动机非线性控制技术研究", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 *

Cited By (36)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107203135A (en) * 2017-06-08 2017-09-26 东莞中子科学中心 A kind of adaptive control system feedovered based on dynamical feedback
CN107608215A (en) * 2017-10-18 2018-01-19 齐鲁工业大学 A kind of mapping UAV Attitude control method based on RBF neural
WO2019144337A1 (en) * 2018-01-25 2019-08-01 大连理工大学 Deep-learning algorithm-based self-adaptive correction method for full-envelope model of aero-engine
CN108828952B (en) * 2018-08-02 2021-03-05 中国民航大学 Intelligent control method of steering engine electric loading system based on cerebellar neural network
CN108828952A (en) * 2018-08-02 2018-11-16 中国民航大学 Steering engine Electric Loading System intelligent control method based on CMAC Neural Network
CN109308004A (en) * 2018-09-28 2019-02-05 东北大学 Loading procedure servo-control system and method behind a kind of true triaxial rigidity matching peak
CN109308004B (en) * 2018-09-28 2021-08-17 东北大学 Real triaxial rigidity testing machine post-peak loading process servo control system and method
CN110852421A (en) * 2019-11-11 2020-02-28 北京百度网讯科技有限公司 Model generation method and device
CN110852421B (en) * 2019-11-11 2023-01-17 北京百度网讯科技有限公司 Model generation method and device
CN111666648B (en) * 2019-12-04 2022-04-12 江西洪都航空工业集团有限责任公司 Method for simulating dynamic characteristics of aircraft engine
CN111666648A (en) * 2019-12-04 2020-09-15 江西洪都航空工业集团有限责任公司 Method for simulating dynamic characteristics of aircraft engine
CN111046568B (en) * 2019-12-18 2022-09-09 哈尔滨工程大学 Control parameter optimization method based on aeroengine and jet nozzle electromechanical liquid multi-system combined simulation
CN111046568A (en) * 2019-12-18 2020-04-21 哈尔滨工程大学 Control parameter optimization method based on aeroengine and jet nozzle electromechanical liquid multi-system combined simulation
CN111273544A (en) * 2020-04-01 2020-06-12 河海大学常州校区 Radar pitching motion control method based on prediction RBF feedforward compensation type fuzzy PID
CN111273544B (en) * 2020-04-01 2022-11-15 河海大学常州校区 Radar pitching motion control method based on prediction RBF feedforward compensation type fuzzy PID
CN111734533A (en) * 2020-05-28 2020-10-02 北京仿真中心 Turbofan engine-based model prediction method and system
CN111734533B (en) * 2020-05-28 2022-09-23 北京仿真中心 Turbofan engine-based model prediction method and system
CN111738563B (en) * 2020-05-29 2023-05-12 中国航天标准化研究所 Method for detecting process control capacity of aerospace production site
CN111738563A (en) * 2020-05-29 2020-10-02 中国航天标准化研究所 Method for detecting control capability of aerospace production field process
CN111624886B (en) * 2020-06-05 2022-04-08 沈阳航空航天大学 Variable-cycle aero-engine thrust control method based on SARSA
CN111624886A (en) * 2020-06-05 2020-09-04 沈阳航空航天大学 Variable-cycle aero-engine thrust control method based on SARSA
CN111880552A (en) * 2020-07-28 2020-11-03 中国科学院长春光学精密机械与物理研究所 Multi-rotor unmanned aerial vehicle trajectory tracking composite control method
WO2022252206A1 (en) * 2021-06-04 2022-12-08 大连理工大学 Aero-engine surge active control system based on fuzzy switching of controllers
CN113485443A (en) * 2021-08-10 2021-10-08 北京宇系航通科技有限公司 Unmanned aerial vehicle control method based on deep learning, storage medium and equipment
CN113485443B (en) * 2021-08-10 2023-12-22 北京北航天宇长鹰无人机科技有限公司 Unmanned aerial vehicle control method based on deep learning, storage medium and equipment
CN113759699A (en) * 2021-08-27 2021-12-07 北京工业大学 Servo motor rotating speed control method for reducing steady-state error
CN114687870A (en) * 2022-03-03 2022-07-01 沈阳航天新光集团有限公司 Vehicle-mounted power take-off generator set control system and control method
CN114839860A (en) * 2022-04-11 2022-08-02 哈尔滨工程大学 Fuzzy PID fuel injection quantity closed-loop control method based on pressure change monitoring of high-pressure natural gas injector inlet
CN114815584A (en) * 2022-04-11 2022-07-29 哈尔滨工程大学 Circulating air injection amount feedforward PID closed-loop control method and system taking natural gas ejector inlet pressure fluctuation as input
CN114839860B (en) * 2022-04-11 2023-02-17 哈尔滨工程大学 Fuzzy PID (proportion integration differentiation) air injection quantity closed-loop control method based on pressure change monitoring of inlet of high-pressure natural gas ejector
CN114706295B (en) * 2022-06-07 2022-08-26 中国飞机强度研究所 Method for controlling fast time-varying thermal load in thermal test for testing strength of aerospace plane
CN114706295A (en) * 2022-06-07 2022-07-05 中国飞机强度研究所 Method for controlling fast time-varying thermal load in thermal test for testing strength of aerospace plane
CN115236975A (en) * 2022-07-28 2022-10-25 南京工业大学 Missile pneumatic heat ground finite element analysis and IPD nonlinear sliding mode control method
CN115236975B (en) * 2022-07-28 2023-09-26 南京工业大学 Missile aerodynamic thermal ground finite element analysis and IPD nonlinear sliding mode control method
CN117075481A (en) * 2023-10-13 2023-11-17 中国空气动力研究与发展中心高速空气动力研究所 Mach number composite control method for wind tunnel transonic ladder variable attack angle test
CN117075481B (en) * 2023-10-13 2023-12-19 中国空气动力研究与发展中心高速空气动力研究所 Mach number composite control method for wind tunnel transonic ladder variable attack angle test

Similar Documents

Publication Publication Date Title
CN106483850A (en) The Fuzzy Self-adaptive PID method for designing that a kind of aero-engine is feedovered based on RBF neural
WO2019144337A1 (en) Deep-learning algorithm-based self-adaptive correction method for full-envelope model of aero-engine
Syahputra APPLICATION OF NEURO-FUZZY METHOD FOR PREDICTION OF VEHICLE FUEL CONSUMPTION.
CN103306822B (en) Aerial turbofan engine control method based on surge margin estimation model
CN107992648A (en) A kind of Adaptive radial basis function neural network carries out aeroengine thrust algorithm for estimating
CN104102769B (en) Artificial intelligence-based method for establishing real time part level model of turbo shaft engine
Bazazzadeh et al. Gas turbine engine control design using fuzzy logic and neural networks
CN111176263B (en) Online identification method for thrust fault of aircraft based on BP neural network
WO2023168821A1 (en) Reinforcement learning-based optimization control method for aeroengine transition state
CN110579962B (en) Turbofan engine thrust prediction method based on neural network and controller
CN110195660A (en) Aero-engine control device based on depth Q study
CN110221537A (en) Control method, device and aero-engine limitation protection control method, device
CN105404750A (en) Turboshaft engine self-adaptive model building method based on modified Broyden algorithm to solve equation set
Zheng et al. A study of aero-engine control method based on deep reinforcement learning
Jaber et al. A new parameters identification of single area power system based LFC using Segmentation Particle Swarm Optimization (SePSO) algorithm
CN113485443B (en) Unmanned aerial vehicle control method based on deep learning, storage medium and equipment
CN111860791A (en) Aero-engine thrust estimation method and device based on similarity transformation
Yangjing et al. Neural network-based model predictive control with fuzzy-SQP optimization for direct thrust control of turbofan engine
CN110985216B (en) Intelligent multivariable control method for aero-engine with online correction
CN105914752A (en) Pilot node selection method based on clustering by fast search and density peaks
Karami et al. Real-time wind direction estimation using machine learning on operational wind farm data
CN114527654A (en) Turbofan engine direct thrust intelligent control method based on reinforcement learning
Li et al. Visualization of CNN transient voltage classification based on feature recognition and enhancement
Zheng et al. A Research on Aero-engine Control Based on Deep Q Learning
Xiong et al. Fusion Prediction of Aero-engine Performance Parameters based on Cubature Kalman Filter

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication

Application publication date: 20170308

RJ01 Rejection of invention patent application after publication