CN110516394A - Aero-engine steady-state model modeling method based on deep neural network - Google Patents

Aero-engine steady-state model modeling method based on deep neural network Download PDF

Info

Publication number
CN110516394A
CN110516394A CN201910823633.5A CN201910823633A CN110516394A CN 110516394 A CN110516394 A CN 110516394A CN 201910823633 A CN201910823633 A CN 201910823633A CN 110516394 A CN110516394 A CN 110516394A
Authority
CN
China
Prior art keywords
neural network
deep neural
engine
aero
state model
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.)
Withdrawn
Application number
CN201910823633.5A
Other languages
Chinese (zh)
Inventor
郑前钢
金崇文
陈浩颖
汪勇
房娟
项德威
胡忠志
张海波
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.)
Nanjing University of Aeronautics and Astronautics
Original Assignee
Nanjing University of Aeronautics and Astronautics
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 Nanjing University of Aeronautics and Astronautics filed Critical Nanjing University of Aeronautics and Astronautics
Priority to CN201910823633.5A priority Critical patent/CN110516394A/en
Publication of CN110516394A publication Critical patent/CN110516394A/en
Withdrawn legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The aero-engine steady-state model modeling method based on deep neural network that the invention discloses a kind of, aero-engine steady-state model is constructed using deep neural network, the deep neural network is successively to criticize normalized deep neural network, it increases by one batch of normalization layer between adjacent hidden layer, is standardized for the output to previous hidden layer.The present invention carries out the modeling of aero-engine steady-state model using deep neural network, and increases the neural network number of plies by introducing batch normalization layer in deep neural network, improves the capability of fitting of network, and then improve the precision of aero-engine steady-state model.

Description

Aero-engine steady-state model modeling method based on deep neural network
Technical field
The present invention relates to Aeroengine control technology field more particularly to a kind of aero-engine steady-state model modeling sides Method.
Background technique
Aero-engine is the aerothermodynamics system of multivariable, strong nonlinearity and complexity, and safe and stable operation is to hair Motivation control system proposes very high requirement, in order to carry out good control to it may first have to establish a mathematical model. It replaces actual engine to carry out simulation study as controlled device using mathematical model, can so save the experiment warp of a large amount of valuableness Take, issuable accident incontrollable accidents when can also avoid debugging control system with actual engine.In addition, advanced Aeroengine control technology, as model base control, flight/propulsion system performance seeking control, Direct Thrust Control, service life Extend control, emergency flight control, performance recovery etc., based on being all airborne engine real-time model with high precision.
Aeroengine modeling method has very much, and popular at present has component-level model, modified linearized model, branch Vector machine and traditional neural network are held, component-level model is its biggest advantage is that model accuracy is high, generally as simulation object, However its real-time is poor, it is difficult to as airborne model；Modified linearized model real-time is high, but since engine is one strong non- Linear object, thus it is bigger to linearize bring modeling error；It support vector machines and the real-time of traditional neural network and builds Mould precision is between component-level model and inearized model that traditional neural network is easy to fall into local optimum, makes model mistake Fitting, support vector machines generalization ability is strong, but it is dfficult to apply to large sample training data, and engine is multivariable, operation Environment is complicated, degeneration and strong nonlinearity object can occur, thus to establish the airborne model that can apply to big envelope curve, training data It necessarily increases, these all limit support vector machines in the application of aeroengine modeling.
Neural network obtains extensive concern since it can theoretically be fitted arbitrary function.Traditional neural network is general Using three layers, with the increase of the network number of plies, network capability of fitting is more and more stronger, but after network number of plies increase, it may appear that Gradient disappears and gradient explosion phenomenon.With the development of nearest more than ten years nerual network technique, especially mentioned in Hinton G E Out after deep learning-depth confidence neural network, neural network makes a breakthrough in many key technologies, and obtains in engineering Huge application, such as speech recognition, figure identification, target detection and Text region etc..However current deep learning- Rarely has application in terms of aero-engine Steady state modeling in terms of deep neural network.
Summary of the invention
The technical problem to be solved by the present invention is to overcome the deficiencies of the prior art and to provide one kind to be based on deep neural network Aero-engine steady-state model modeling method, the precision of aero-engine steady-state model can be effectively improved.
The present invention specifically uses following technical scheme to solve above-mentioned technical problem:
A kind of aero-engine steady-state model modeling method based on deep neural network, is constructed using deep neural network Aero-engine steady-state model, the deep neural network are successively to criticize normalized deep neural network, adjacent implicit Increase by one batch of normalization layer between layer, is standardized for the output to previous hidden layer.
Preferably, the standardization is specific as follows:
Wherein,For the output after standardization, ε is the positive integer of numerical value very little,To enter batch normalization layer Neural network output before, μBWithThe respectively mean value and variance of sample data set, γ and β are two learning parameters.
It is further preferred that the modeling method the following steps are included:
Step 1, the training data for obtaining aero-engine steady-state model；
Step 2 determines the structure for successively criticizing normalized deep neural network；
Step 3 carries out forward calculation to successively criticizing normalized deep neural network, obtains loss function value；
Step 4 successively criticizes normalized deep neural network gradient using back-propagation algorithm calculating, and updates weight；
Step 5, judgement successively criticize whether normalized deep neural network restrains, be to export steady-state model, otherwise after Continuous iteration, return step 3.
Preferably, the aeroplane engine is obtained by engine test experiment or/and engine non-linear components grade model The training data of machine steady-state model.
Preferably, the aero-engine steady-state model is with flying height, Mach number, fuel flow, jet pipe venturi face Product, phoenix fan guide vane angle and compressor guide vane angle be mode input amount, with engine oil consumption rate, installed thrust, fan propeller revolving speed, Compressor rotor revolving speed, fan surge margin, compressor surge nargin and high-pressure turbine inlet temperature are model output.
Compared with prior art, technical solution of the present invention has the advantages that
The present invention carries out the modeling of aero-engine steady-state model using deep neural network, and by depth nerve net Batch normalization layer is introduced in network to increase the neural network number of plies, improves the capability of fitting of network, and then it is steady to improve aero-engine The precision of states model.
Detailed description of the invention
Fig. 1 is five layers of neural network structure schematic diagram；
Fig. 2 is the structural schematic diagram for successively criticizing normalized deep neural network；
Fig. 3 is data profile；
Fig. 4 is Sigmod curve graph；
Fig. 5 is backpropagation schematic illustration；
Fig. 6 is the opposite training error of deep neural network training；
Fig. 7 is the opposite training error of three layers of BP neural network training；
Fig. 8 is the opposite test error of deep neural network training；
Fig. 9 is the opposite test error of three layers of BP neural network training.
Specific embodiment
The case where being difficult to improve present invention is generally directed to traditional aero-engine steady-state process modeling method precision proposes one Aero-engine steady-state model modeling method of the kind based on deep neural network, this method use the depth for successively criticizing normalization method Neural network increases by one batch of normalization layer between adjacent hidden layer, marks for the output to previous hidden layer Quasi-ization processing, so that the network number of plies of the modeling method proposed is high, capability of fitting is strong, and then it is steady to improve aero-engine The modeling accuracy of states model.
The present invention proposes the aero-engine steady-state model modeling method based on deep neural network, mainly includes following Several steps: step 1, the training data for obtaining aero-engine steady-state model；
Step 2 determines the structure for successively criticizing normalized deep neural network；
Step 3 carries out forward calculation to successively criticizing normalized deep neural network, obtains loss function value；
Step 4 successively criticizes normalized deep neural network gradient using back-propagation algorithm calculating, and updates weight；
Step 5, judgement successively criticize whether normalized deep neural network restrains, be to export steady-state model, otherwise after Continuous iteration, return step 3.
Engine steady state data are engine parameter when engine stabilizer is run, and data can pass through engine test reality It tests or/and engine non-linear components grade model obtains, it is generally non-thread by engine at present since test run experimental cost is high Property component-level model obtains engine steady state data.
By taking five layers of neural network as an example, basic structure is as shown in Figure 1.W in figureiAnd biI=1,2,3,4 is respectively weight And biasing, J are loss function.To w1And b1Local derviation is sought, is obtained:
Because of σ ' (hi) derivative be smaller than 0.25, work as wiWhen≤1, w2σ′(h2)≤0.25, this illustrates that the number of plies is more, Gradient journey index reduces, which is known as gradient disappearance；Furthermore it is assumed that wi>=100, σ ' (h2)=0.1, then w2σ′(h2) >=10, This will appear again, and with the increase of the number of plies, gradient journey index increases, which is known as gradient spilling.
In order to overcome the problems, such as gradient disappearance, the present invention normalizes (BNBatchNormalization) using layer-by-layer batch, it Increase by one BN layers between two hidden layers, it can effectively prevent gradient and disappear and gradient overflow problem, network knot Structure as shown in Figure 2
In neural network weights initialisation, often allowing weight to meet mean value is zero, the Gaussian Profile that variance is 1, with This simultaneously, input data is standardized, and output data is normalized or standardization.By mapping, training, often One layer of data distribution is all changed, and widely different, this leads to the distributional difference of each layer of neural network weight very Greatly, each layer of neural network learning rate is often identical when training, this greatly reduces the convergence rate of network.If logarithm According to whitening operation is carried out, the principle is as follows shown in figure, and Fig. 3 a is training data distribution map, and it is inclined to can be seen that data distribution in figure It from Gaussian Profile, carries out subtracting mean value, after the operations such as decorrelation, data as shown in Figure 3b is obtained, so that data fit Gaussian Profile, to accelerate the learning rate of neural network.
There are many kinds of whitening operations, and there are commonly PCA albefactions, it is that data is allowed to meet 0 mean value, unit variance and weak phase It closes, however, albefaction is worthless, is primarily due to whitening operation and the operation such as needs to calculate covariance matrix, invert, it is computationally intensive, Moreover, whitening operation can not necessarily be led when backpropagation, then using batch normalization, standard is carried out to each hidden layer node Change processing, it is after weight product, before activation primitive, it is assumed that neural network forward-propagating is the defeated of l i-th of node of layer It is outWherein j ∈ χk, k represents kth group training dataset, then has
WhereinFor the output after normalization, ε is the positive integer of numerical value very little,For the nerve before BN layers of entrance Network output, μBWithRespectively mean value and variance, calculation formula are as follows
hl=σ (Wl-1hl-1+bl-1) (4)
However, the ability to express of network can be reduced if only carrying out batch normalizing operation.As shown in figure 4, if activation letter When number is sigmoid, data are limited in zero mean unit variance, then being only equivalent to the linear portion using activation primitive Point, and the non-linear partial of two sides is seldom related to, this will obviously reduce the ability to express of network.
For this purpose, the present invention increases the two learning parameters of γ and β, to keep the ability to express of network, expression formula again It is as follows:
The μ that above formula acquiresBWithIt is to be acquired under minimum lot size (Min-batch), and should be theoretically entire data The mean value and variance of collection.
The training method of deep neural network is also using back-propagation algorithm, and the network parameter for needing to update is W, b, γ, β are updated as follows using gradient descent method:
Gradient is acquired using back-propagation algorithm, principle as shown in fig. 5, it is assumed thatFor
Assuming that δlFor
Wherein l=nnet,nnet-2,…,2
For l=nnet-1,nnet- 2 ..., 2Have
Therefore the gradient of network parameter is obtained are as follows:
In addition to BN layers, other layers of derivation formula are identical in MGDNN.
In order to verify the validity and advance of aero-engine Steady state modeling method proposed by the invention, with small bypass ratio Aero-engine component-level model is simulation object, establishes the aeroplane engine for performance optimizing based on proposition method of the present invention The airborne model of machine, and compared with MGD-NN, MGD-NN uses minimum lot size gradient descent method (MGD, mini-batch Gradient descent) network is trained, overcome traditional neural network to be not applied for the determination of big-sample data, is Make deep neural network use and large sample training, is equally trained using MGD method, depth hereinafter referred to as proposed in this paper Degree neural network steady-state model is known as BN-MGD-DNN.It is selected by cross validation brush, the network structure for obtaining BN-MGD-DNN is [6,10,15,15,10,7], the network structure of MGD-NN are [6,40,7], and minimum training sample set is 3000 in MGD algorithm, just Then changing constant is 10-6.
For aircraft in cruise, flying height H and Mach number Ma are slowly varying, and engine control amount is in addition to fuel oil WfbAnd tail Nozzle throat area A8, influence of the guide vane angle angle of fan and compressor to oil consumption rate is also very big, therefore, herein with H, Ma, Wfb、A8, phoenix fan guide vane angle αfWith compressor guide vane angle αcFor mode input amount, engine oil consumption rate Sfc, installed thrust Fin, fan Rotor speed Nf, compressor rotor revolving speed Nc, fan surge margin Smf, compressor surge nargin SmcWith high-pressure turbine inlet temperature T4For model output, the prediction model for constructing engine parameter is as follows:
Y=fBN-MGD-DNN(x) (22)
Wherein
Since neural network is similar to non-linear interpolation device, in interpolated value, precision is high, and in extrapolated value, precision is low, therefore Selected training sample includes to input the maximum value and minimum value of parameter, and in order to avoid over-fitting, train sample as much as possible Should be more as much as possible, for subsonic speed and supersonic cruise, H value range 9~13km, Ma choose 0.7~1.5, WfbVariation Range is to change with PLA and Ma, A8Variation range is the jet pipe throat area A from design point8,dsTo 1.3A8,ds, αfAnd αc's Variation range is -3 ° to 3 °.Choosing training sample set is 3726498, and test sample collection chooses 7536
The opposite training error of BN-MGD-DNN and MGD-NN is set forth in Fig. 6-9, as can be seen from the figure BN-MGD- The error of DNN is substantially 3% hereinafter, meet required precision, and its training precision is apparently higher than MGD-NN, especially Sfc、Nf、 SmfAnd Smc, one times or so of training precision ratio MGD-NN high.Fig. 6 and 7 gives the training phase of BN-MGD-DNN and MGD-NN To error, it can be seen from the figure that the test error of BN-MGD-DNN is in addition to SmfAnd SmcWithin 2%, its other precision are all Within 1%, meet required precision, and from Fig. 8 and 9 as can be seen that deep neural network ratio of precision tradition BP neural network There are larger raising, especially fan and compressor rotor revolving speed and surge margin, it is stronger that this illustrates that BN-MGD-DNN has Generalization ability.
Table 1 gives the average opposite test error peace of BN-MGD-DNN and MGD-NN with respect to training error, with MGD-NN is compared, and BN-MGD-DNN has higher training precision and measuring accuracy.Wherein set forth herein BN-MGD-DNN modelings The S of methodfc、Nf、Nc、Fin、T4,SmfAnd SmcAverage trained relative error respectively reduced 1.4 than MGD-NN, 2.17, 2.0, it 1.3,1.13,2.4 and 2.8 times, for the average opposite test error especially relevant with model generalization performance, respectively reduces 1.75,2.0,2.3,1.3,1.3,2.3 and 3.3 times.
Table 2 provides the data storage of MGD-NN and BN-MGD-DNN, computation complexity, mean test time, can from table To find out that the algorithm complexity of the two is low, data storage capacity is small, mean test time is short, all meet airborne requirement.
Wherein the data storage capacity of MGD-NN is 567 (weight 520 (6 × 40+40 × 7)+47 biases (40+7))；BN- The data storage capacity of MGD-DNN is 940
The computation complexity of MGD-NN is 614 (multiplyings 520 (6 × 40+40 × 7)+add operation 47 (40+7)+swash Function 47 (40+7) living)；
The data storage capacity of BN-MGD-DNN is the 940 ((6 × 10+10 × 15+15 × 15+15 × 10+10 of multiplying 712 × 7+10+15+15+10+7)+division arithmetic 57 (10+15+15+10+7)+57 (10+15+15+10+7+10+15+ of add operation 15+10+7)+subtraction 57 (10+15+15+10+7)+activation primitive 57 (10+15+15+10+7))
Two program execution environments are all are as follows: operating system Windows 7Ultimate with Service Pack 1 (x64)；Processor (CPU) is Intel (R) Core (TM) i5-4590h, its dominant frequency is 3.30GHz, and memory (RAM) is 8G, The software of operation is MATLAB2016a, and performance optimizing mode simulation environment is identical as this, is not being illustrated below, can be with from table Find out that MGD-NN the and BN-MGD-DNN testing time is respectively 0.067 millisecond and 0.223 millisecond.
The average opposite test of table 1 and training error table
2 MGD-NN and BN-MGD-DNN algorithm comparison of table

Claims (5)

1. a kind of aero-engine steady-state model modeling method based on deep neural network is constructed using deep neural network and is navigated Empty Stable status engine model, which is characterized in that the deep neural network is successively to criticize normalized deep neural network, Increase by one batch of normalization layer between adjacent hidden layer, is standardized for the output to previous hidden layer.
2. aero-engine steady-state model modeling method as described in claim 1, which is characterized in that the standardization is specific It is as follows:
Wherein,For the output after standardization, ε is the positive integer of numerical value very little,To enter before batch normalization layer Neural network output, μBWithThe respectively mean value and variance of sample data set, γ and β are two learning parameters.
3. aero-engine steady-state model modeling method as claimed in claim 2, which is characterized in that the modeling method include with Lower step:
Step 1, the training data for obtaining aero-engine steady-state model；
Step 2 determines the structure for successively criticizing normalized deep neural network；
Step 3 carries out forward calculation to successively criticizing normalized deep neural network, obtains loss function value；
Step 4 successively criticizes normalized deep neural network gradient using back-propagation algorithm calculating, and updates weight；
Step 5, judgement successively criticize whether normalized deep neural network restrains, and are to export steady-state model, otherwise continue to change Generation, return step 3.
4. aero-engine steady-state model modeling method as claimed in claim 3, which is characterized in that tested by engine test Or/and engine non-linear components grade model obtains the training data of the aero-engine steady-state model.
5. the aero-engine steady-state model modeling method as described in any one of Claims 1 to 4, which is characterized in that the aviation Stable status engine model is led with flying height, Mach number, fuel flow, jet pipe throat area, phoenix fan guide vane angle and compressor Leaf angle is mode input amount, with engine oil consumption rate, installed thrust, fan propeller revolving speed, compressor rotor revolving speed, fan surge Nargin, compressor surge nargin and high-pressure turbine inlet temperature are model output.
CN201910823633.5A 2019-09-02 2019-09-02 Aero-engine steady-state model modeling method based on deep neural network Withdrawn CN110516394A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910823633.5A CN110516394A (en) 2019-09-02 2019-09-02 Aero-engine steady-state model modeling method based on deep neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910823633.5A CN110516394A (en) 2019-09-02 2019-09-02 Aero-engine steady-state model modeling method based on deep neural network

Publications (1)

Publication Number Publication Date
CN110516394A true CN110516394A (en) 2019-11-29

Family

ID=68630337

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910823633.5A Withdrawn CN110516394A (en) 2019-09-02 2019-09-02 Aero-engine steady-state model modeling method based on deep neural network

Country Status (1)

Country Link
CN (1) CN110516394A (en)

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111486009A (en) * 2020-04-23 2020-08-04 南京航空航天大学 Aero-engine control method and device based on deep reinforcement learning
CN111914461A (en) * 2020-09-08 2020-11-10 北京航空航天大学 Intelligent assessment method for one-dimensional cold efficiency of turbine guide vane
CN113282004A (en) * 2021-05-20 2021-08-20 南京航空航天大学 Neural network-based aeroengine linear variable parameter model establishing method
CN113485117A (en) * 2021-07-28 2021-10-08 沈阳航空航天大学 Multivariable reinforcement learning control method for aircraft engine based on input and output information
CN113741170A (en) * 2021-08-17 2021-12-03 南京航空航天大学 Aero-engine direct thrust inverse control method based on deep neural network
CN113804446A (en) * 2020-06-11 2021-12-17 卓品智能科技无锡有限公司 Diesel engine performance prediction method based on convolutional neural network

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109598220A (en) * 2018-11-26 2019-04-09 山东大学 A kind of demographic method based on the polynary multiple dimensioned convolution of input

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109598220A (en) * 2018-11-26 2019-04-09 山东大学 A kind of demographic method based on the polynary multiple dimensioned convolution of input

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
ZHENG,QIANGANG 等: "Aero-Engine On-Board Model Based on Batch Normalize Deep Neural Network", 《IEEE ACCESS》 *

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111486009A (en) * 2020-04-23 2020-08-04 南京航空航天大学 Aero-engine control method and device based on deep reinforcement learning
CN113804446A (en) * 2020-06-11 2021-12-17 卓品智能科技无锡有限公司 Diesel engine performance prediction method based on convolutional neural network
CN111914461A (en) * 2020-09-08 2020-11-10 北京航空航天大学 Intelligent assessment method for one-dimensional cold efficiency of turbine guide vane
CN113282004A (en) * 2021-05-20 2021-08-20 南京航空航天大学 Neural network-based aeroengine linear variable parameter model establishing method
CN113282004B (en) * 2021-05-20 2022-06-10 南京航空航天大学 Neural network-based aeroengine linear variable parameter model establishing method
CN113485117A (en) * 2021-07-28 2021-10-08 沈阳航空航天大学 Multivariable reinforcement learning control method for aircraft engine based on input and output information
CN113485117B (en) * 2021-07-28 2024-03-15 沈阳航空航天大学 Multi-variable reinforcement learning control method for aeroengine based on input and output information
CN113741170A (en) * 2021-08-17 2021-12-03 南京航空航天大学 Aero-engine direct thrust inverse control method based on deep neural network

Similar Documents

Publication Publication Date Title
CN110516394A (en) Aero-engine steady-state model modeling method based on deep neural network
US11823057B2 (en) Intelligent control method for dynamic neural network-based variable cycle engine
US5649064A (en) System and method for modeling the flow performance features of an object
CN112859898B (en) Aircraft trajectory prediction method based on two-channel bidirectional neural network
Liang et al. Multi-objective robust airfoil optimization based on non-uniform rational B-spline (NURBS) representation
CN115618733B (en) Multi-scale hybrid attention mechanism modeling method for predicting remaining service life of aircraft engine
CN112149233A (en) Aero-engine dynamic thrust estimation method based on echo state network
CN106980262A (en) Self-adaptive flight device robust control method based on Kernel recursive least square algorithm
Körpe et al. Aerodynamic optimization of a UAV wing subject to weight, geometric, root bending moment, and performance constraints
CN110516391A (en) A kind of aero-engine dynamic model modeling method neural network based
Kumar et al. GPR-based novel approach for non-linear aerodynamic modelling from flight data
Asouti et al. PCA-enhanced metamodel-assisted evolutionary algorithms for aerodynamic optimization
Qiao et al. Far-field drag decomposition using hybrid formulas and vorticity based area sensors
Zheng et al. On-board real-time optimization control for turbo-fan engine life extending
Angelini et al. Identification of losses in turbomachinery with machine learning
Teng et al. Generative adversarial surrogate modeling framework for aerospace engineering structural system reliability design
CN112116101B (en) Aeroengine fault diagnosis method based on group reduction kernel extreme learning machine
CN109115446A (en) Based on transonic wind tunnel wind speed accuracy control method and system
CN110362065A (en) A kind of method for diagnosing status of aero-engine anti-asthma control system
CN109177101A (en) A kind of injection molding machine batch process fault detection method
Veldman et al. Vistrafs: a simulation method for strongly interacting viscous transonic flow
Thurman et al. Comparison of Prediction Modeling Methodologies for Aeroacoustic Characterization of Hovering sUAS Rotors
Ivanov et al. Fast Compressor Map Computation by Utilizing Support Vector Machine and Response Surface Approximation
CN113741170B (en) Aero-engine direct thrust inverse control method based on deep neural network
Wang et al. Incremental support vector machine learning method for aircraft event recognition

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
WW01 Invention patent application withdrawn after publication

Application publication date: 20191129

WW01 Invention patent application withdrawn after publication