CN108375474B - A kind of aero-engine transition state critical performance parameters prediction technique - Google Patents
A kind of aero-engine transition state critical performance parameters prediction technique Download PDFInfo
- Publication number
- CN108375474B CN108375474B CN201810075189.9A CN201810075189A CN108375474B CN 108375474 B CN108375474 B CN 108375474B CN 201810075189 A CN201810075189 A CN 201810075189A CN 108375474 B CN108375474 B CN 108375474B
- Authority
- CN
- China
- Prior art keywords
- data
- relative error
- performance parameter
- parameters
- para
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 50
- 230000007704 transition Effects 0.000 title claims abstract description 16
- 238000012360 testing method Methods 0.000 claims abstract description 22
- 238000012549 training Methods 0.000 claims abstract description 15
- 238000005096 rolling process Methods 0.000 claims abstract description 8
- 238000005516 engineering process Methods 0.000 claims abstract description 4
- 239000002245 particle Substances 0.000 claims description 24
- 230000008569 process Effects 0.000 claims description 18
- 238000012706 support-vector machine Methods 0.000 claims description 18
- 230000000694 effects Effects 0.000 claims description 12
- 238000004422 calculation algorithm Methods 0.000 claims description 10
- 238000012952 Resampling Methods 0.000 claims description 8
- 230000006870 function Effects 0.000 claims description 7
- 238000005457 optimization Methods 0.000 claims description 7
- 238000012545 processing Methods 0.000 claims description 7
- 208000034530 PLAA-associated neurodevelopmental disease Diseases 0.000 claims description 6
- 238000010219 correlation analysis Methods 0.000 claims description 6
- 238000011156 evaluation Methods 0.000 claims description 6
- 238000005070 sampling Methods 0.000 claims description 6
- 230000001133 acceleration Effects 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 5
- 238000004140 cleaning Methods 0.000 claims description 5
- 238000010606 normalization Methods 0.000 claims description 5
- 238000004458 analytical method Methods 0.000 claims description 3
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 230000010354 integration Effects 0.000 claims description 3
- 230000007246 mechanism Effects 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 238000012800 visualization Methods 0.000 claims description 3
- 230000009977 dual effect Effects 0.000 claims description 2
- 238000002790 cross-validation Methods 0.000 claims 1
- 239000012634 fragment Substances 0.000 claims 1
- 230000004927 fusion Effects 0.000 abstract description 4
- 230000008859 change Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000012423 maintenance Methods 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000007619 statistical method Methods 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 102100024452 DNA-directed RNA polymerase III subunit RPC1 Human genes 0.000 description 1
- 101000689002 Homo sapiens DNA-directed RNA polymerase III subunit RPC1 Proteins 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 230000000630 rising effect Effects 0.000 description 1
- 238000010845 search algorithm Methods 0.000 description 1
- 230000001052 transient effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M15/00—Testing of engines
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B64—AIRCRAFT; AVIATION; COSMONAUTICS
- B64F—GROUND OR AIRCRAFT-CARRIER-DECK INSTALLATIONS SPECIALLY ADAPTED FOR USE IN CONNECTION WITH AIRCRAFT; DESIGNING, MANUFACTURING, ASSEMBLING, CLEANING, MAINTAINING OR REPAIRING AIRCRAFT, NOT OTHERWISE PROVIDED FOR; HANDLING, TRANSPORTING, TESTING OR INSPECTING AIRCRAFT COMPONENTS, NOT OTHERWISE PROVIDED FOR
- B64F5/00—Designing, manufacturing, assembling, cleaning, maintaining or repairing aircraft, not otherwise provided for; Handling, transporting, testing or inspecting aircraft components, not otherwise provided for
- B64F5/60—Testing or inspecting aircraft components or systems
Landscapes
- Engineering & Computer Science (AREA)
- Manufacturing & Machinery (AREA)
- Transportation (AREA)
- Aviation & Aerospace Engineering (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Control Of Turbines (AREA)
Abstract
The invention belongs to aero-engine performance parameter prediction technical fields, provide a kind of aero-engine transition state critical performance parameters prediction technique.Provided aero-engine transition state bench test data, which are studied, using certain first establishes training dataset and test data set；Based on the thought of information fusion, delivery temperature is predicted and analyzed using parametric joint；And rolling study is carried out using moving window technology, to predict from practical implementation angle parameters such as rotational speed of lower pressure turbine rotor, the delivery temperatures of engine.
Description
Technical Field
The invention belongs to the technical field of prediction of performance parameters of an aero-engine, and particularly relates to a prediction method of transition state key performance parameters of the aero-engine.
Background
The aircraft engine, as the most central component of an aircraft, has an operating state that directly determines the stability and safety of the entire aircraft. While the performance of the aircraft engine is greatly improved, the structure is more complicated, and the state monitoring and maintenance are increasingly difficult. The performance of the transition state of the engine directly relates to the performance of the aircraft such as takeoff, accelerated flight, maneuvering flight and the like, so that the requirements on the rapidity, the stability, the safety and the reliability of the transition state of the aircraft engine are extremely high. The performance parameters of the engine can reflect the health condition of the engine, so that the important performance parameters of the transition state of the engine are predicted, the working state of the engine can be mastered in real time, the current engine state monitoring and fault prediction capabilities are effectively improved, and the purpose of improving the reliability and safety of the operation of the engine is achieved.
At present, many scholars at home and abroad develop work in the aspect of prediction of aeroengine performance parameters, and the work mainly comprises a model-based method, a statistical-based method, a regression-based method and a machine learning-based method. The model-based method is complex in calculation, and the problems of iteration unconvergence and the like can occur during real-time calculation; the statistical method mainly comprises the steps of carrying out statistical analysis on faults, maintenance records and the like of the engine; regression-based methods sometimes do not necessarily have a significant linear or other functional relationship between the variables, and the model is difficult to select. The method based on machine learning has very strong nonlinear mapping capability, can update and evolve by self, and has short training time and high learning speed.
Disclosure of Invention
Aiming at the defects in the prior art, the invention provides a prediction method of transition state performance parameters of an aero-engine based on a support vector machine, wherein a training data set and a test data set are established by using test data of a transition state rack of the aero-engine provided by a certain research institute; based on the idea of information fusion, predicting and analyzing the exhaust temperature by adopting parameter combination; and rolling learning is carried out by utilizing a moving window technology, and effective real-time prediction is realized from the perspective of engineering application.
The technical scheme of the invention is as follows:
a method for predicting key performance parameters of an aircraft engine in a transition state comprises the following steps:
first, preprocessing the aeroengine performance parameter data
(1) The performance parameter data of the aircraft engine comprises the throttle lever angle PLA and the low-pressure rotor rotating speed n1High-pressure rotor speed n2Ambient pressure p0Outlet pressure p of high-pressure compressor31Oil pressure pfCompressor outlet temperature t1Exhaust temperature EGT, lead angle α29 sets of parameters;
(2) data integration: the aeroengine performance parameter data comprises a plurality of aeroengine test run process field acquisition data, the aeroengine test run process field acquisition data are combined and stored uniformly, and an aeroengine performance parameter data warehouse is established;
(3) resampling: analyzing the aeronautical launch performance parameter data, and adopting a linear resampling method to resample the aeronautical launch performance parameter data for facilitating later rolling prediction due to different sampling time intervals;
(4) normalization: normalization processing is carried out on the aviation performance parameter data after resampling processing, and the data are converted into data within a certain range, so that magnitude difference among all dimensional data is eliminated, and large prediction error caused by large magnitude difference of input and output data is avoided. The max-min method is used, and the conversion form is as follows:
x＝(xnor-xmin)/(xmax-xmin)
in the formula, xnorFor the data sequence to be normalized, xminIs the smallest number, x, in the data sequencemaxIs the maximum number in the data sequence;
(5) data screening and cleaning: carrying out visualization processing on the normalized aviation performance parameter data, and clustering and cleaning an acceleration curve;
secondly, analyzing the correlation of the performance parameter data of the aircraft engine
The early aircraft engine performance parameter prediction method adopts single parameter prediction, so that the prediction result cannot be ensured, and the data resource is greatly wasted. Therefore, based on the idea of information fusion, certain parameter is predicted and analyzed by parameter combination. If the correlation between the fused information is not great, the obtained result is often unsatisfactory, and even the fusion prediction effect is inferior to the single-parameter prediction effect. The method analyzes direct influence parameters of the rotating speed of the low-pressure rotor by combining the process of the aeroengine mechanism while performing correlation analysis by adopting a gray level correlation method.
The gray level correlation analysis steps are as follows: first, a reference sequence is selected and recorded as g 0:
g0＝{g0(j)|j＝1,2,...p}＝(g0(1),g0(2)...,g0(p))
then selecting comparison sequence, and recording it as gi：
gi＝{gi(j)|j＝1,2,...p}＝(gi(1),gi(2)...,gi(p)),i＝1,2...,q
Respectively calculating the mean value of the index correlation coefficients of each evaluation object to reflect each evaluation comparison sequence giWith reference sequence g0And the association relationship is called as the association degree, and is recorded as:
wherein the correlation coefficient ξi(j) The following calculations were made:
finally, the throttle lever angle PLA and the oil pressure p are selectedfLead angle α2Three parameters as input, low-pressure rotor speed n1As an output quantity;
thirdly, constructing a training database
In order to reflect the time-varying characteristics of the engine transition state performance parameters, a training database is constructed by adopting a moving window technology, and a schematic diagram is shown in fig. 2. The figure has two data windows, the solid line frame is the input data window, the dashed line frame is the output data window (also called the prediction window), the widths of the two windows are respectivelyIs TDAnd TP(ii) a The input data window and the output data window together with a step size TMIs moved to the right (T)MFor moving step size) to obtain dynamic process data segments at different times, thereby obtaining corresponding input-output data vector pairs.
An input-output vector pair corresponding to the kth data window is defined as { X (T)k),Y(Tk) Suppose the selected performance parameter is Para1、Para2、...、ParanAnd for the parameter ParaiAnd predicting, then:
X(Tk)＝[Tk,x(Tk),x(Tk-1τ),…,x(Tk-mτ)]
x(Tk)＝[Para1(Tk),…,Paran(Tk)]
m＝TD/τ
Y(Tk)＝Parai(Tk+TP)
wherein, TkIs the time corresponding to the right end of the input data window, τ is the discretization step, m is the number of equal parts of the input data window, and the input vector X (T)k) From TkAnd (m +1) discrete values of process variables at sampling moments covered by a data window closed interval, and outputting a vector Y (T)k) For outputting the right end of the data window corresponding to the time (T)k+TP) Of the variable Para to be predictediThe actual value of (c);
fourthly, constructing a prediction model based on support vector regression
The step is mainly composed of two parts, firstly, the rotation speed n of the low-pressure rotor is controlled by using a support vector machine1And performing rolling learning prediction, and optimizing the support vector machine through a group intelligent algorithm so as to construct a prediction model.
(1) Low-voltage rotor speed n by using support vector machine1Performing rolling learning predictions
In the real classification decision, it is often difficult to determine a proper kernel function so that the training sample can be linearly separable in the feature space, and even if the training sample is linearly separable, it is difficult to judge that the result is not caused by overfitting. To alleviate this problem, the support vector machine is allowed to classify errors on some samples.
I.e. introducing the concept of "soft spacing", allowing some samples not to satisfy the constraint: y isi(ωTxi+b)≥1。
A commonly used soft-space support vector machine is:
the constraint conditions need to be satisfied:
yi(ωTxi+b)≥1-ξi
ξi≥0，i＝1，...，m
at this time, the dual problem constraint of the objective function convex quadratic programming optimization can be transformed into:
considering that the number of Gaussian Radial Basis Function (RBF) kernel parameters is less, the parameter optimization is convenient to follow, and meanwhile, the model is relatively stable. SVM selects RBF kernel functionWhere σ is a kernel parameter whose size affects the shape of the kernel function, and the larger σ, the smaller the nonlinear efficiency and the less sensitive to noise. And x, xiIs a sample.
(2) Optimizing support vector machine parameters by group intelligence algorithm
And optimizing parameters such as C, sigma and the like in the support vector machine by adopting a particle swarm algorithm. The particle swarm algorithm is to initialize a group of particles in a search space, and each particle is a potential optimal solution of the extremum optimization problem. Three indexes of position, speed and fitness value are used for representing the characteristics of the particles, and the fitness value is used for measuring the quality of the particles.
Assume that in the d-dimensional search space, a population T composed of n particles is (T ═ T)1，T2，...，Tn) Wherein the ith particle represents a d-dimensional vector Ti＝(ti1，ti2，...，tid)。
Let the velocity of the ith particle be Vi＝(Vi1，Vi2，...，Vid)TWith an individual extremum of Pi＝(Pi1，Pi2，...，Pid)TThe population extremum of the population is Pg＝(Pg1，Pg2，...，Pgd)T. In each iteration, the updated formula for the velocity and position of the particle can be expressed as:
where w is the inertial weight, s is the number of current iterations, VijIs the particle velocity, acceleration factor c1，c2Not less than 0, random number r1，r2∈[0，1]。
Fitness values in particle features use cross-validated classification accuracy as an indicator.
And fifthly, testing the prediction model by using the test set to evaluate the prediction effect.
The main evaluation indexes include:
(1) average relative error (MRE) and Total average relative error of 70 batches (A _ MRE)
The average relative error is formulated as
The total average relative error of 70 batches is
(2) Maximum relative error of 70 batches (MAX _ MAXRE)
In the trial run process, the early-stage numerical value of the EGT is small, and even if the prediction effect is good, the relative error is still large, so that the analysis of the maximum relative error is divided into two parts, namely, the maximum relative error in the whole process is solved, and the maximum relative error from the 30 th moment is solved.
The maximum relative error is given by
The maximum relative error of 50 batches is
MAX_MAXRE1＝max(MAXRE1j),
j＝1,2,…,N
MAX_MAXRE2＝max(MAXRE2j),
j＝1,2,…,N
(3) Root Mean Square Error (RMSE) and the total mean root mean square error (A _ RMSE) of 70 batches
Root mean square error of
The total mean root mean square error in 70 batches is
Wherein,the predicted value of the current batch j at the ith moment is obtained; x is the number ofijIs the true value; n is the total number of batches, namely 70; n is the time series length of the current batch j.
The invention has the beneficial effects that: the invention provides a support vector machine-based prediction method for key performance parameters of an aircraft engine in a transition state, which optimizes the support vector machine by adopting an intelligent algorithm, thereby predicting parameters such as the rotating speed of a low-pressure rotor, the exhaust temperature and the like of the engine from the perspective of practical engineering application.
Drawings
FIG. 1 is a flow chart of a prediction model building process for a transient state performance parameter of an aircraft engine.
Fig. 2 is a schematic diagram of a moving window.
Fig. 3 is a graph (trend graph) of the forecasting effect of the low-pressure rotor speed (N1) in 3 test runs.
Fig. 4 is a prediction effect graph (error graph) of the low-pressure rotor speed (N1) in 3 trial batches.
Fig. 5 is the N1 prediction error (lead 2S) for the 70 test lots.
Detailed Description
The following further describes a specific embodiment of the present invention with reference to the drawings and technical solutions.
The data used is a set of transitional stage bench test data 180 for a particular type of aircraft engine provided by a research institute in the country.
First, preprocessing the aviation performance parameter data
(1) The data category mainly comprises the throttle lever angle PLA and the low-pressure rotor rotating speed n1High-pressure rotor speed n2Ambient pressure p0Outlet pressure p of high-pressure compressor31Oil pressure pfCompressor outlet temperature t1Exhaust temperature EGT, lead angle α2There are 9 sets of parameters.
(2) Data integration: and combining and uniformly storing data (such as excel and txt) in a plurality of data sources to establish a data warehouse of the aviation performance parameters.
(3) Resampling: the analysis is performed on the collected data due to the unequal time intervals. Therefore, in order to facilitate subsequent rolling prediction, data is first resampled. The specific operation steps are as follows, the sampling frequency which is newly drawn up is inserted into the time sequence of the original data as the interpolation by adopting the interpolation method, and then the number of the original data between every two rated sampling points is counted. If only one original data is contained, the original data is taken as the data corresponding to the sampling point; if the time point data comprises two original data, averaging the two original data to obtain the data corresponding to the time point; and if the data does not contain the original data, taking the average value of the corresponding data of the previous time point and the next time point of the time point in the rated time sequence as the data of the time point.
(4) Normalization: carrying out normalization processing on the data, converting the data into data in a certain range, and using a maximum and minimum method, wherein the conversion form is as follows:
x＝(xnor-xmin)/(xmax-xmin)
in the formula, xnorFor the data sequence to be normalized, xminIs the smallest number, x, in the data sequencemaxIs the maximum number in the data sequence;
(5) data screening and cleaning: and carrying out visualization processing on the data so as to simply cluster and clear the acceleration curve.
Second, parameter correlation analysis
The method analyzes direct influence parameters of the rotating speed of the low-pressure rotor by combining an aircraft engine mechanism process while performing correlation analysis by adopting a gray scale correlation method.
Finally, the throttle lever angle PLA and the oil pressure p are selectedfLead angle α2Three parameters as input quantities, and low-pressure rotor speed n1As an output. The experimental results can also verify that the three parameters are fused and then subjected to combined prediction to achieve better effect.
Thirdly, constructing a training database
Randomly selected 110 groups of data as training data set and 70 groups of data as testing data set. The relevant parameter of the experimental training database is set to TD＝0.5s、TP＝2.0s、TM0.1s and τ 0.1 s. That is, input data of
The output data is
Y(t+2)＝[N1(t+2)]
Wherein x (t) ═ pla (t), pf (t), α2(t)]And L is the time length of each test run process.
That is, the forecast is carried out in 2s ahead, the forecast rolls to the right side of the time shaft for 0.1s after one time, the forecast is continued, and the like. Since the time length of each test run is different, the number of training samples is gradually reduced from 110 groups as time goes on.
The fourth step: building a prediction model
In addition to the support vector machine based on ion swarm optimization (PSO-SVM) model presented herein, we also used a Kernel Extreme Learning Machine (KELM) model and a least squares support vector machine based on quantum swarm optimization (QPSO-LSSM) model as a control. Wherein KELM uses RBF kernel function, parameter C160. And considering the requirement on the calculation efficiency in engineering practice, the QPSO-LSSVM sets the group size of the search algorithm to be 3 and the maximum iteration number to be 7.
The fifth step: result discussion vs. indicators
The method provided by the invention and the two comparison methods are adopted. And respectively training the models by using the training sets, and analyzing and comparing the prediction effects of the models by using the test sets. Four indicators of the total average relative error (A _ MRE), the maximum relative error (MAX _ MAXRE1 and MAX _ MAXRE2) and the total average root mean square error (A _ RMSE) of the 70 batches were calculated, respectively. The model predicted N1 index is shown in table 1.
TABLE 1 indexes of prediction results of different algorithms on transition state N1 of aircraft engine
Meanwhile, as shown in fig. 3, for the PSO-SVM model used, a trend effect graph (3 typical batches with large errors at the same time) is used for predicting three batches in the test set. The solid line represents an observed value, and the dotted line represents a predicted value. It can be seen from the figure that the rising and falling trends of the predicted curve and the original observed curve are approximately the same, and at the inflection point of the curve (i.e. at different power switching moments), the predicted curve can follow the slope change of the original curve, which also means that the method proposed herein can reasonably predict the configuration change of the engine 2s ahead. Fig. 4 shows the relative error of three batches.
As shown in FIG. 5, the abscissa of the graph represents the test lot number, which is 70 in total. The ordinate is the average of the relative error of the low pressure rotor speed for each batch. Except that the relative error of a few batches is more than + 1.4%, the error of the other batches is basically distributed in the interval [ 0.8%, 1.2% ], namely the method achieves better prediction effect on each test sample.
Claims (1)
1. A prediction method for key performance parameters of an aircraft engine in a transition state is characterized by comprising the following steps:
first, preprocessing the aeroengine performance parameter data
(1) The performance parameter data of the aircraft engine comprises the throttle lever angle PLA and the low-pressure rotor rotating speed n1High-pressure rotor speed n2Ambient pressure p0Outlet pressure p of high-pressure compressor31Oil pressure pfCompressor outlet temperature t1Exhaust temperature EGT, lead angle α29 sets of parameters;
(2) data integration: the aeroengine performance parameter data comprises a plurality of aeroengine test run process field acquisition data, the aeroengine test run process field acquisition data are combined and stored uniformly, and an aeroengine performance parameter data warehouse is established;
(3) resampling: analyzing the aeronautical launch performance parameter data, and resampling the aeronautical launch performance parameter data by adopting a linear resampling method;
(4) normalization: normalizing the aircraft performance parameter data after resampling processing by using a maximum and minimum method, wherein the conversion form is as follows:
x＝(xnor-xmin)/(xmax-xmin)
in the formula, xnorFor the data sequence to be normalized, xminIs the smallest number, x, in the data sequencemaxIs the maximum number in the data sequence;
(5) data screening and cleaning: carrying out visualization processing on the normalized aviation performance parameter data, and clustering and cleaning an acceleration curve;
secondly, analyzing the correlation of the performance parameter data of the aircraft engine
The method is characterized in that correlation analysis is carried out by adopting a gray level correlation method, and simultaneously, direct influence parameters of the rotating speed of the low-pressure rotor are analyzed by combining an aircraft engine mechanism process;
the gray level correlation analysis steps are as follows: firstly, a reference sequence is selected and recorded as g0：
g0＝{g0(j)|j＝1,2,...p}＝(g0(1),g0(2)...,g0(p))
Then selecting comparison sequence, and recording it as gi：
gi＝{gi(j)|j＝1,2,...p}＝(gi(1),gi(2)...,gi(p)),i＝1,2...,q
Respectively calculating the mean value of the index correlation coefficients of each evaluation object to reflect each evaluation comparison sequence giWith reference sequence g0And the association relationship is called as the association degree, and is recorded as:
wherein the correlation coefficient ξi(j) The following calculations were made:
finally, the throttle lever angle PLA and the oil pressure p are selectedfLead angle α2Three parameters as input, low-pressure rotor speed n1As an output quantity;
thirdly, constructing a training database
In order to reflect the time-varying characteristic of the engine transition state performance parameters, a moving window technology is adopted to construct a training database, two data windows, an input data window and an output data window are arranged, and the widths of the two windows are respectively TDAnd TP(ii) a The input data window and the output data window together with a step size TMThe speed of the process is moved to the right, and dynamic process data fragments of different time are obtained, so that corresponding input-output data vector pairs are obtained;
an input-output vector pair corresponding to the kth data window is defined as { X (T)k),Y(Tk) Suppose the selected performance parameter is Para1、Para2、…、ParanAnd for the parameter ParaiAnd predicting, then:
X(Tk)＝[Tk,x(Tk),x(Tk-1τ),…,x(Tk-mτ)]
x(Tk)＝[Para1(Tk),…,Paran(Tk)]
m＝TD/τ
Y(Tk)＝Parai(Tk+TP)
wherein, TkIs the time corresponding to the right end of the input data window, τ is the discretization step, m is the number of equal parts of the input data window, and the input vector X (T)k) From TkAnd data window closed interval covered (m)+1) discrete values of the process variable at the sampling instant, the output vector Y (T)k) For outputting the right end of the data window corresponding to the time (T)k+TP) Of the variable Para to be predictediThe actual value of (c);
fourthly, constructing a prediction model based on support vector regression
Mainly comprises two parts, firstly, a support vector machine is utilized to rotate the low-pressure rotor at a speed n1Performing rolling learning prediction, and optimizing the support vector machine through a group intelligent algorithm to construct a prediction model;
(1) using support vector machine SVM to measure low-voltage rotor speed n1Performing rolling learning predictions
The "soft interval" concept was introduced, allowing some samples not to satisfy the constraint: y isi(ωTxi+b)≥1；
The soft interval support vector machine is adopted as follows:
the constraint conditions need to be satisfied:
yi(ωTxi+b)≥1-ξi
ξi≥0，i＝1，...，m
at this time, the dual problem constraint condition of the objective function convex quadratic programming optimization is transformed into:
SVM selects RBF kernel functionWherein sigma is a kernel parameter, the size of the kernel parameter can influence the shape of a kernel function, and the larger the sigma is, the smaller the nonlinear efficiency is, and the less sensitive to noise is; x, xiIs a sample;
(2) optimizing support vector machine parameters by group intelligence algorithm
Optimizing C and sigma parameters in the support vector machine by adopting a particle swarm algorithm; firstly, initializing a group of particles in a search space by a particle swarm algorithm, wherein each particle is a potential optimal solution of an extremum optimization problem; the characteristics of the particles are represented by three indexes, namely position, speed and fitness value, and the quality of the particles is measured by the fitness value;
assume that in the d-dimensional search space, a population T composed of n particles is (T ═ T)1，T2，...，Tn) Wherein the ith particle represents a d-dimensional vector Ti＝(ti1，ti2，...，tid)；
Let the velocity of the ith particle be Vi＝(Vi1，Vi2，...，Vid)TWith an individual extremum of Pi＝(Pi1，Pi2，...，Pid)TThe population extremum of the population is Pg＝(Pg1，Pg2，...，Pgd)T(ii) a In each iteration, the update formula for the velocity and position of the particle is represented as:
where w is the inertial weight, s is the number of current iterations, VijIs the particle velocity, acceleration factor c1，c2Not less than 0, random number r1，r2∈[0，1]；
The fitness value in the particle features uses the classification accuracy of cross validation as an index;
fifthly, testing the prediction model by using the test set to evaluate the prediction effect
The main evaluation indexes include:
(1) average relative error MRE and total average relative error A _ MRE of 70 batches
The average relative error is formulated as
The total average relative error of 70 batches is
(2) Maximum relative error MAX _ MAXRE among 70 batches
In the trial run process, the early-stage numerical value of the EGT is small, and even if the prediction effect is good, the relative error is still large, so that the analysis of the maximum relative error is divided into two parts, namely, the maximum relative error in the whole process is solved, and the maximum relative error from the 30 th moment is solved;
the maximum relative error is given by
The maximum relative error of 50 batches is
MAX_MAXRE1＝max(MAXRE1j),
j＝1,2,…,N
MAX_MAXRE2＝max(MAXRE2j),
j＝1,2,…,N
(3) Root mean square error RMSE and the total mean root mean square error A _ RMSE of 70 batches
Root mean square error of
The total mean root mean square error in 70 batches is
Wherein,the predicted value of the current batch j at the ith moment is obtained; x is the number ofijIs the true value; n is the total number of batches, namely 70; n is the time series length of the current batch j.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810075189.9A CN108375474B (en) | 2018-01-26 | 2018-01-26 | A kind of aero-engine transition state critical performance parameters prediction technique |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810075189.9A CN108375474B (en) | 2018-01-26 | 2018-01-26 | A kind of aero-engine transition state critical performance parameters prediction technique |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108375474A CN108375474A (en) | 2018-08-07 |
| CN108375474B true CN108375474B (en) | 2019-06-21 |
Family
ID=63016885
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810075189.9A Active CN108375474B (en) | 2018-01-26 | 2018-01-26 | A kind of aero-engine transition state critical performance parameters prediction technique |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108375474B (en) |
Families Citing this family (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109611217B (en) * | 2018-11-07 | 2020-12-11 | 大连理工大学 | Design method for optimizing transition state control law of aircraft engine |
| WO2020093264A1 (en) * | 2018-11-07 | 2020-05-14 | 大连理工大学 | Design method for optimizing aero-engine transition state control law |
| CN109799803B (en) * | 2018-12-11 | 2020-06-16 | 大连理工大学 | LFT-based aeroengine sensor and actuator fault diagnosis method |
| CN111114825A (en) * | 2019-12-24 | 2020-05-08 | 中国航空工业集团公司西安飞机设计研究所 | Intelligent filter for airplane and filter element detection method |
| CN111219257B (en) * | 2020-01-07 | 2022-07-22 | 大连理工大学 | Turbofan engine direct data drive control method based on adaptive enhancement algorithm |
| CN111361759B (en) * | 2020-03-02 | 2023-02-03 | 哈尔滨工业大学 | Airplane auxiliary power device on-wing residual life prediction method based on hybrid model |
| CN113408560B (en) * | 2020-03-17 | 2024-04-16 | 联合汽车电子有限公司 | Engine test data classification method, electronic device, and readable storage medium |
| CN111649951B (en) * | 2020-04-15 | 2021-10-15 | 中国电子产品可靠性与环境试验研究所（（工业和信息化部电子第五研究所）（中国赛宝实验室）） | Method and device for detecting faults of aircraft engine, computer equipment and storage medium |
| CN111679574B (en) * | 2020-05-13 | 2021-05-07 | 大连理工大学 | Variable-cycle engine transition state optimization method based on large-scale global optimization technology |
| CN112610339B (en) * | 2021-01-13 | 2021-12-28 | 南京航空航天大学 | Variable cycle engine parameter estimation method based on proper amount of information fusion convolutional neural network |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| DE102011090151A1 (en) * | 2011-12-30 | 2013-07-04 | Robert Bosch Gmbh | Method for forecasting a rotational speed of a drive shaft of an internal combustion engine |
| CN104951851A (en) * | 2015-07-08 | 2015-09-30 | 华侨大学 | Wind turbine state prediction model establishing method based on grey relation-regression SVM (support vector machine) |
| CN105740984A (en) * | 2016-02-01 | 2016-07-06 | 北京理工大学 | Product concept performance evaluation method based on performance prediction |
| CN107368913A (en) * | 2017-06-15 | 2017-11-21 | 中国汽车技术研究中心 | Oil consumption prediction method based on least square support vector machine |
| US10127497B2 (en) * | 2014-10-14 | 2018-11-13 | Microsoft Technology Licensing, Llc | Interface engine for efficient machine learning |
-
2018
- 2018-01-26 CN CN201810075189.9A patent/CN108375474B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| DE102011090151A1 (en) * | 2011-12-30 | 2013-07-04 | Robert Bosch Gmbh | Method for forecasting a rotational speed of a drive shaft of an internal combustion engine |
| US10127497B2 (en) * | 2014-10-14 | 2018-11-13 | Microsoft Technology Licensing, Llc | Interface engine for efficient machine learning |
| CN104951851A (en) * | 2015-07-08 | 2015-09-30 | 华侨大学 | Wind turbine state prediction model establishing method based on grey relation-regression SVM (support vector machine) |
| CN105740984A (en) * | 2016-02-01 | 2016-07-06 | 北京理工大学 | Product concept performance evaluation method based on performance prediction |
| CN107368913A (en) * | 2017-06-15 | 2017-11-21 | 中国汽车技术研究中心 | Oil consumption prediction method based on least square support vector machine |
Non-Patent Citations (3)
| Title |
|---|
| 基于支持向量机和粒子群算法的信息网络安全态势复合预测模型;高昆仑 等;《电网技术》;20110430;第35卷(第4期);第176-182页 |
| 基于模糊信息粒化和优化SVM的航空发动机性能趋势预测;李艳军 等;《航空动力学报》;20171231;第32卷(第12期);第3022-3030页 |
| 基于神经网络的涡扇发动机中间稳态参数模型;潘鹏飞 等;《工程与试验》;20151231;第55卷(第4期);第9-12页 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108375474A (en) | 2018-08-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108375474B (en) | A kind of aero-engine transition state critical performance parameters prediction technique | |
| US11124317B2 (en) | Method for prediction of key performance parameters of aero-engine in transition condition | |
| CN109766583B (en) | Aircraft engine life prediction method based on unlabeled, unbalanced and initial value uncertain data | |
| CN110361176B (en) | Intelligent fault diagnosis method based on multitask feature sharing neural network | |
| WO2020000248A1 (en) | Space reconstruction based method for predicting key performance parameters of transition state acceleration process of aircraft engine | |
| WO2024087128A1 (en) | Multi-scale hybrid attention mechanism modeling method for predicting remaining useful life of aero engine | |
| CN115048874B (en) | Aircraft design parameter estimation method based on machine learning | |
| CN111680875B (en) | Unmanned aerial vehicle state risk fuzzy comprehensive evaluation method based on probability baseline model | |
| CN106778846A (en) | A kind of method for forecasting based on SVMs | |
| CN116448419A (en) | Zero sample bearing fault diagnosis method based on depth model high-dimensional parameter multi-target efficient optimization | |
| Li et al. | Aero-engine exhaust gas temperature prediction based on LightGBM optimized by improved bat algorithm | |
| CN115062528A (en) | Prediction method for industrial process time sequence data | |
| CN116227367B (en) | Back pressure prediction model construction method, back pressure prediction method and back pressure prediction device of direct air cooling system | |
| CN116307139A (en) | Wind power ultra-short-term prediction method for optimizing and improving extreme learning machine | |
| CN112416913B (en) | GWO-BP algorithm-based aircraft fuel system state missing value supplementing method | |
| Cui et al. | Prediction of Aeroengine Remaining Useful Life Based on SE-BiLSTM | |
| Shaowu et al. | Prediction of exhaust gas temperature margin based on LSSVR | |
| CN112365022A (en) | Engine bearing fault prediction method based on multiple stages | |
| Zhang et al. | Health status characterization and RUL prediction method of escalator bearing based on VMD-GRU | |
| CN117852411B (en) | Modeling design gas compressor pneumatic performance prediction method and system based on neural network | |
| CN114330819B (en) | Wind turbine generator free inflow wind speed prediction method | |
| CN113256018B (en) | Wind power ultra-short term probability prediction method based on conditional quantile regression model | |
| CN114841000B (en) | Soft measurement modeling method based on modal common feature separation | |
| CN113591384B (en) | Cement finished product specific surface area prediction method based on gating convolution network | |
| Fenglei et al. | Prediction of engine total pressure distortion in improved cascaded forward network |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |