CN112765899B - Turboshaft engine multi-target performance prediction method based on Bayesian classifier chain - Google Patents

Turboshaft engine multi-target performance prediction method based on Bayesian classifier chain Download PDF

Info

Publication number
CN112765899B
CN112765899B CN202110140394.0A CN202110140394A CN112765899B CN 112765899 B CN112765899 B CN 112765899B CN 202110140394 A CN202110140394 A CN 202110140394A CN 112765899 B CN112765899 B CN 112765899B
Authority
CN
China
Prior art keywords
power
section temperature
target variable
probability
turboshaft engine
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
Application number
CN202110140394.0A
Other languages
Chinese (zh)
Other versions
CN112765899A (en
Inventor
蔡志强
韩思杰
王宇航
司书宾
张帅
叶正梗
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.)
Northwestern Polytechnical University
Original Assignee
Northwestern Polytechnical University
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 Northwestern Polytechnical University filed Critical Northwestern Polytechnical University
Priority to CN202110140394.0A priority Critical patent/CN112765899B/en
Publication of CN112765899A publication Critical patent/CN112765899A/en
Application granted granted Critical
Publication of CN112765899B publication Critical patent/CN112765899B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • G06F18/24155—Bayesian classification
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/10—Geometric CAD
    • G06F30/15—Vehicle, aircraft or watercraft design
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/10—Geometric CAD
    • G06F30/17—Mechanical parametric or variational design
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/04—Inference or reasoning models
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00—Details relating to CAD techniques
    • G06F2111/08—Probabilistic or stochastic CAD
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/06—Power analysis or power optimisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/08—Thermal analysis or thermal optimisation

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Geometry (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Computer Hardware Design (AREA)
  • Artificial Intelligence (AREA)
  • Data Mining & Analysis (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • Software Systems (AREA)
  • Mathematical Analysis (AREA)
  • Computational Mathematics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Aviation & Aerospace Engineering (AREA)
  • Automation & Control Theory (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computational Linguistics (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Probability & Statistics with Applications (AREA)
  • Medical Informatics (AREA)
  • Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)

Abstract

The invention discloses a multi-target performance prediction method of a turboshaft engine based on a Bayes classifier chain, which connects Bayes classifiers according to the relation among target variables, establishes a chain performance prediction model, and predicts the qualification probability of a plurality of performance target variables of the turboshaft engine simultaneously through posterior probability reasoning after the attribute variable state of the turboshaft engine to be predicted is given. Meanwhile, in order to ensure the accuracy of the result, different connection sequences of the target variable are considered, the prediction results of all models are averaged, and the final performance prediction result of the target variable is obtained, so that the production can be guided, and the factory yield of the turboshaft engine is improved.

Description

Turboshaft engine multi-target performance prediction method based on Bayesian classifier chain
Technical Field
The invention belongs to the technical field of aero-engines, and particularly relates to a method for predicting multi-target performance of an engine.
Background
The turboshaft engine is a highly complex precision thermal machine, is generally used as a power source of a helicopter, and has extremely high requirements on the manufacturing level. Generally, a qualified turboshaft engine has two performance parameter indicators to be considered: power and critical cross-sectional temperature. In order to ensure that the helicopter can be always provided with sufficient and stable power, the power of the engine has the lowest qualified limit; meanwhile, in order to ensure the working life of the engine and the safety of the helicopter, the index of the critical section temperature has the highest qualified limit. However, in actual production, the manufactured engine is difficult to meet the qualified requirements of two indexes by one trial run, and multiple trial runs are often tried after the engine is reassembled, so that the labor and material costs are greatly increased. The performance of the turboshaft engine can be accurately and effectively predicted, so that the risk can be predicted in advance, and the production link can be guided, so that the ex-factory yield of the turboshaft engine is improved.
For the performance prediction of an aircraft engine, at the present stage, there are two main methods: model-based methods and data-based methods. The model-based method carries out performance prediction work by constructing an accurate engine mathematical model, and the application of the method needs to be established on the basis of deep research on the structure and the principle of an engine. Furthermore, the method is overly dependent on the accuracy of the mathematical model, and the model prediction results are susceptible to noise and modeling uncertainty. Data-based methods do not require prior knowledge of the engine system, but rather predict its performance directly from collected engine data, of which machine learning algorithms are a typical representative. With the rise of artificial intelligence and the accumulation of a large amount of engine data in recent years, data-based methods attract more and more researchers and are gradually becoming the first solution in the field of aircraft engine performance prediction.
The Beijing university of aerospace school, 2008, 34 (3) discloses a method for predicting performance parameters of an aircraft engine, which comprises the steps of decomposing original data of the engine into a plurality of groups of subsequences on different scales by utilizing wavelet transformation, respectively selecting an autoregressive moving average model or a summation autoregressive moving average model for prediction according to the characteristics of each subsequence, and finally synthesizing all prediction results to obtain a total prediction result. However, the current data-based prediction methods have the following disadvantages: modeling prediction is inefficient, prediction of multiple performance parameter indexes cannot be considered simultaneously, and specific performance qualified prediction probability cannot be given.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides a multi-target performance prediction method of a turboshaft engine based on a Bayes classifier chain, which connects Bayes classifiers according to the relation among target variables, establishes a chain-shaped performance prediction model, and can predict the qualification probability of a plurality of performance target variables of the turboshaft engine to be predicted at the same time through posterior probability reasoning after the attribute variable state of the turboshaft engine to be predicted is given. Meanwhile, in order to ensure the accuracy of the result, different connection sequences of the target variable are considered, the prediction results of all models are averaged, and the final performance prediction result of the target variable is obtained, so that the production can be guided, and the factory yield of the turboshaft engine is improved.
The technical scheme adopted by the invention for solving the technical problem comprises the following steps:
step 1: collecting performance parameter data and design parameter data of a turboshaft engine;
acquiring performance parameter data of the turboshaft engine by testing a plurality of turboshaft engines, wherein the performance parameter data comprises measurement data of power P and key section temperature T during the test of the turboshaft engine;
obtaining design parameter data of a turboshaft engine, wherein the design parameter data comprises the sizes of N different parts in the turboshaft engine;
and 2, step: determining a target variable and an attribute variable, and setting a qualified condition of the target variable; discretizing each variable to obtain a probability distribution table and a conditional probability table of the target variable and conditional probability tables of various states of the attribute variables under different target variables; the method comprises the following specific steps:
step 2-1: setting the power P and the critical section temperature T of the turboshaft engine as target variables, and setting the sizes of N different parts in the turboshaft engine as attribute variables, wherein the x is represented as 1 ,x 2 ,…,x i ,…,x N ，i＝1,2,…,N；
Step 2-2: setting the minimum qualified condition of the power P as a KW; setting the maximum qualified condition of the critical section temperature T as b ℃, wherein a and b respectively represent two normal constants;
step 2-3: dividing the power P into two sections by taking the minimum qualified condition of the power P as a boundary, and counting the probability distribution table P of the power P in the two sections by the data obtained in the step 1 r (P); dividing the critical section temperature T into two sections by taking the maximum qualified condition of the critical section temperature T as a boundary, and counting the probability distribution table P of the critical section temperature T in the two sections according to the data obtained in the step 1 r (T)；
The data statistical power P and the key section temperature T obtained in the step 1 belong to a conditional probability table P of different segments in a priori manner r (P | T) and P r (T|P)；
Step 2-4: each attribute variable x 1 ,x 2 ,…,x N The method is characterized in that the method is divided into e sections by adopting an equal frequency method, the e sections are respectively expressed as states 0,1, \8230, e-1,0,1, \8230, and e-1 represents the state of an attribute variable and does not represent the actual value of the attribute variable; for the ith attribute variable, the threshold values adopted for dividing the attribute variable into e segments are respectively
Figure GDA0003853552770000021
Figure GDA0003853552770000022
In correspondence with the state 0, the state,
Figure GDA0003853552770000023
in correspondence with the state 1, the state of the mobile terminal,
Figure GDA0003853552770000024
corresponding to the state 2, and so on,
Figure GDA0003853552770000025
corresponding to state e-1;
step 2-5: counting N attribute variables by the data obtained in the step 1, and taking the power P and the key section temperature T as prior numbers to belong to a conditional probability table P in different states r (x i P) and P r (x i |T)；
And step 3: obtaining a performance prediction result of the turboshaft engine on the target variable power P and the key section temperature T under a Bayes classifier chain performance prediction model based on a naive Bayes classifier and taking the target variable power P as a chain head through posterior probability reasoning; the specific process is as follows:
step 3-1: the Bayes classifier chain performance prediction model with the target variable power P as the chain head comprises two classifiers, wherein the first classifier is used for predicting the target variable power P, the target variable power P is used as a father node, and each attribute variable x is 1 ,x 2 ,…,x N As child nodes; the second classifier is used for predicting the key section temperature T of the target variable, taking the key section temperature T of the target variable as a father node and taking each attribute variable x 1 ,x 2 ,…,x N And the target variable power P as child nodes;
step 3-2: setting an attribute variable x 1 ,x 2 ,…,x N Respectively, is y 1 ,y 2 ,…,y N ，y 1 ,y 2 ,…,y N ∈[0,1,…,e-1](ii) a Predicting power PHELpass (PHELpass) according to a first classifier of a Bayes classifier chain performance prediction model taking target variable power P as a chain head>a) The probability of (d);
the posterior probability of power Pqualified (P > a) is then:
Figure GDA0003853552770000031
the posterior probability of unqualified power P (P ≦ a) is:
Figure GDA0003853552770000032
according to the basic properties of conditional probabilities, there are:
Figure GDA0003853552770000033
substituting the numerical value in the conditional probability table obtained in the step 2 to solve P r (x 1 ＝y 1 ,x 2 ＝y 2 ,…,x N ＝y N ) Finally, calculating to obtain power PHELpass (PHELpass)>a) Has a probability of P r (P>a|x 1 ＝y 1 ,x 2 ＝y 2 ,…,x N ＝y N )＝m％；
If m% is less than 50%, judging that the power P of the turboshaft engine to be predicted is unqualified under a Bayes classifier chain performance prediction model taking the target variable power P as a chain head; if m% is more than or equal to 50%, judging that the power P of the turboshaft engine to be predicted is qualified under a Bayes classifier chain performance prediction model taking the target variable power P as a chain head;
step 3-3: predicting the probability of the qualified temperature T (T is less than or equal to b) of the key section according to a second classifier of a Bayes classifier chain performance prediction model taking the target variable power P as a chain head;
the posterior probability of the key section temperature T qualified (T is less than or equal to b) is as follows:
Figure GDA0003853552770000041
wherein, when m% <50%, it is [ < no ]; when m% is more than or equal to 50%, the signal is more than or equal to the number;
the posterior probability of the critical section temperature T disqualification (T > b) is:
Figure GDA0003853552770000042
according to the basic properties of conditional probabilities, there are:
Figure GDA0003853552770000043
substituting the numerical value in the conditional probability table obtained in the step 2 to solve P r (x 1 ＝y 1 ,x 2 ＝y 2 ,…,x N ＝y N P ^ a), the probability of obtaining the critical section temperature Tpass (T ≦ b) finally calculated as P r (T≤b|x 1 ＝y 1 ,x 2 ＝y 2 ,…,x N ＝y N ,P⊙a)＝n％；
If n% <50%, judging that the critical section temperature T of the turboshaft engine to be predicted is unqualified under a Bayes classifier chain performance prediction model taking the target variable power P as a chain head; if n% is more than or equal to 50%, judging that the critical section temperature T of the turboshaft engine to be predicted is qualified under a Bayes classifier chain performance prediction model taking the target variable power P as a chain head;
and 4, step 4: the performance prediction results of the turboshaft engine on the target variable power P and the key section temperature T under a Bayes classifier chain performance prediction model based on a naive Bayes classifier and taking the target variable key section temperature T as a chain head are obtained through posterior probability reasoning; the specific process is as follows:
step 4-1: the Bayes classifier chain performance prediction model taking the target variable key section temperature T as the chain head comprises two classifiers, wherein the first classifier is used for predicting the target variable key section temperature T, the target variable key section temperature T is used as a father node, and each attribute variable x is subjected to prediction 1 ,x 2 ,…,x N As child nodes; the second classifier takes the target variable power P as a parent node and takes each attribute variable x as the prediction of the target variable power P 1 ,x 2 ,…,x N And the target variable key section temperature T is taken as a child node;
step 4-2: predicting the probability of the qualification (T is less than or equal to b) of the key section temperature T according to a first classifier of a Bayes classifier chain performance prediction model taking the target variable key section temperature T as a chain head;
the posterior probability of the key section temperature T qualified (T is less than or equal to b) is as follows:
Figure GDA0003853552770000051
the posterior probability of the critical section temperature T disqualification (T > b) is:
Figure GDA0003853552770000052
according to the basic properties of conditional probabilities, there are:
Figure GDA0003853552770000053
substituting the numerical value in the conditional probability table obtained in the step 2 to solve P r (x 1 ＝y 1 ,x 2 ＝y 2 ,…,x N ＝y N ) Finally, the probability that the critical section temperature T is qualified (T is less than or equal to b) is calculated and obtained as P r (T≤b|x 1 ＝y 1 ,x 2 ＝y 2 ,…,x N ＝y N )＝s％；
If s% <50%, judging that the critical section temperature T of the turboshaft engine to be predicted is unqualified under a Bayes classifier chain performance prediction model taking the target variable critical section temperature T as a chain head; if s% is more than or equal to 50%, judging that the critical section temperature T of the turboshaft engine to be predicted is qualified under a Bayes classifier chain performance prediction model taking the target variable critical section temperature T as a chain head;
step 4-3: predicting the probability of power P qualification (P > a) according to a second classifier of a Bayes classifier chain performance prediction model taking the target variable key section temperature T as a chain head;
the posterior probability of power Pqualified (P > a) is then:
Figure GDA0003853552770000061
wherein, when s%<When the content of the organic acid is 50 percent,
Figure GDA0003853552770000062
is composed of>Number; when the s% is more than or equal to 50%,
Figure GDA0003853552770000063
is no more than number;
the posterior probability of unqualified power P (P ≦ a) is:
Figure GDA0003853552770000064
according to the basic properties of conditional probabilities, there are:
Figure GDA0003853552770000065
substituting the numerical value in the conditional probability table obtained in the step 2 to solve
Figure GDA0003853552770000066
Figure GDA0003853552770000067
Finally, the power P qualified is obtained through calculation (P)>a) Has a probability of
Figure GDA0003853552770000068
If the r% is less than 50%, judging that the power P of the turboshaft engine to be predicted is unqualified under a Bayes classifier chain performance prediction model taking the target variable key section temperature T as a chain head; if the r% is more than or equal to 50%, judging that the power P of the turboshaft engine to be predicted is qualified under a Bayes classifier chain performance prediction model taking the key section temperature T of the target variable as a chain head;
and 5: averaging the qualification probability prediction results of the power P and the key section temperature T obtained in the step 3 and the step 4 under a Bayes classifier chain performance prediction model with the target variable power P and the target variable key section temperature T as chain heads respectively to finally obtain: the final qualification probability prediction result of the power P of the turboshaft engine to be predicted is
Figure GDA0003853552770000069
The final qualification probability prediction result of the critical section temperature T is
Figure GDA00038535527700000610
Step 6: if it is not
Figure GDA00038535527700000611
Finally judging that the power P of the turboshaft engine to be predicted is unqualified; if it is not
Figure GDA00038535527700000612
Finally judging that the power P of the turboshaft engine to be predicted is qualified;
if it is not
Figure GDA0003853552770000071
Finally judging that the critical section temperature T of the turboshaft engine to be predicted is unqualified; if it is used
Figure GDA0003853552770000072
And finally judging that the critical section temperature T of the turboshaft engine to be predicted is qualified.
Preferably, said e =3.
The invention has the following beneficial effects:
1. the method disclosed by the invention completes the construction of a Bayes classifier chain performance prediction model based on a naive Bayes classifier, performs performance prediction on the turboshaft engine through posterior probability reasoning, and is simple and efficient in modeling and prediction processes.
2. The method can predict two performance parameter indexes at the same time, considers the possible correlation between the two indexes, and better accords with the actual situation of engine performance prediction compared with a single-target prediction method.
3. The method can effectively utilize the test data of the turboshaft engine to predict the performance of the turboshaft engine, and the prediction accuracy is ensured by averaging the prediction results of different models. In addition, the final given result is the specific performance qualified prediction probability, which is beneficial to scientific quantitative evaluation of the performance of the turboshaft engine and has important significance for improving the overall reliability of the turboshaft engine.
Drawings
FIG. 1 is a schematic flow diagram of the process of the present invention.
Fig. 2 is a model for predicting the performance of a chain of a bayesian classifier with a target variable power P as a chain head according to an embodiment of the present invention.
FIG. 3 is a Bayesian classifier chain performance prediction model with a target variable critical section temperature T as a chain head according to an embodiment of the present invention.
Detailed Description
The invention is further illustrated with reference to the following figures and examples.
Step 1: collecting performance parameter data and design parameter data of the turboshaft engine;
acquiring performance parameter data of the turboshaft engine by testing a plurality of turboshaft engines, wherein the performance parameter data comprises measurement data of power P and key section temperature T during the test of the turboshaft engine, and the data type is continuous;
obtaining design parameter data of a turboshaft engine, wherein the design parameter data comprises the sizes of N different parts in the turboshaft engine, and the data type is continuous; these design parameters are generally considered to have an effect on the performance of the engine and may be determined by the experience of the skilled person or by reference to relevant literature;
step 2: determining a target variable and an attribute variable, and setting a qualified condition of the target variable; discretizing each variable to obtain a probability distribution table and a conditional probability table of the target variable and conditional probability tables of various states of the attribute variables under different target variables; the method comprises the following specific steps:
step 2-1: setting the power P and the critical section temperature T of the turboshaft engine as target variables, and setting the sizes of N different parts in the turboshaft engine as attribute variables, wherein the x is represented as 1 ,x 2 ,…,x i ,…,x N ，i＝1,2,…,N；
Step 2-2: setting the minimum qualified condition of the power P as a KW; setting the maximum qualified condition of the temperature T of the key section as b ℃, wherein a and b respectively represent two normal constants;
step 2-3: dividing the power P into two sections by taking the minimum qualified condition of the power P as a boundary, and counting the probability distribution table P of the power P in the two sections by the data obtained in the step 1 r (P)；Dividing the critical section temperature T into two sections by taking the maximum qualified condition of the critical section temperature T as a boundary, and counting the probability distribution table P of the critical section temperature T in the two sections according to the data obtained in the step 1 r (T)；
The data statistical power P and the key section temperature T acquired in the step 1 are mutually prior conditional probability tables P belonging to different sections r (P | T) and P r (T|P)；
Step 2-4: each attribute variable x 1 ,x 2 ,…,x N The method comprises the steps of respectively dividing the sections into sections e by adopting an equal frequency method, and respectively representing the sections e into states 0,1, \8230, e-1,0,1, \8230, wherein the e-1 represents the state of an attribute variable and does not represent the actual value of the attribute variable; for the ith attribute variable, the threshold values adopted for dividing the ith attribute variable into e segments are respectively
Figure GDA0003853552770000081
Figure GDA0003853552770000082
In correspondence with the state 0, the state,
Figure GDA0003853552770000083
in correspondence with the state 1, the state of the mobile terminal,
Figure GDA0003853552770000084
corresponding to the state 2, and so on,
Figure GDA0003853552770000085
corresponding to state e-1;
step 2-5: counting N attribute variables by the data obtained in the step 1, and taking the power P and the key section temperature T as prior numbers to belong to a conditional probability table P in different states r (x i P) and P r (x i |T)；
And step 3: obtaining a performance prediction result of the turboshaft engine on the target variable power P and the key section temperature T under a Bayes classifier chain performance prediction model based on a naive Bayes classifier and taking the target variable power P as a chain head through posterior probability reasoning; the specific process is as follows:
step 3-1: taking target variable power P as chainThe first Bayes classifier chain performance prediction model comprises two classifiers, wherein the first classifier is used for predicting target variable power P, the target variable power P is used as a father node, and each attribute variable x is used as a parent node 1 ,x 2 ,…,x N As child nodes; the second classifier is used for predicting the key section temperature T of the target variable, taking the key section temperature T of the target variable as a father node and taking each attribute variable x 1 ,x 2 ,…,x N And the target variable power P as child nodes;
step 3-2: setting an attribute variable x 1 ,x 2 ,…,x N Respectively, is y 1 ,y 2 ,…,y N ，y 1 ,y 2 ,…,y N ∈[0,1,…,e-1](ii) a Predicting power Pqualified (Pqualified) according to a first classifier of a Bayes classifier chain performance prediction model taking target variable power P as a chain head>a) The probability of (d);
the a posteriori probability of power phir (P > a) is then:
Figure GDA0003853552770000091
the posterior probability of unqualified power P (P ≦ a) is:
Figure GDA0003853552770000092
according to the basic properties of conditional probabilities, there are:
Figure GDA0003853552770000093
substituting the numerical value in the conditional probability table obtained in the step 2 to solve P r (x 1 ＝y 1 ,x 2 ＝y 2 ,…,x N ＝y N ) Finally, calculating to obtain power PHELpass (PHELpass)>a) Has a probability of P r (P>a|x 1 ＝y 1 ,x 2 ＝y 2 ,…,x N ＝y N )＝m％；
If m% is less than 50%, judging that the power P of the turboshaft engine to be predicted is unqualified under a Bayes classifier chain performance prediction model taking the target variable power P as a chain head; if m% is more than or equal to 50%, judging that the power P of the turboshaft engine to be predicted is qualified under a Bayes classifier chain performance prediction model taking the target variable power P as a chain head;
step 3-3: predicting the probability of the qualified temperature T (T is less than or equal to b) of the key section according to a second classifier of a Bayes classifier chain performance prediction model taking the target variable power P as a chain head;
the posterior probability of the key section temperature T qualified (T is less than or equal to b) is as follows:
Figure GDA0003853552770000094
wherein, when m% <50%, it is [ < no ]; when m% is more than or equal to 50%, the signal is more than or equal to the mark;
the posterior probability of the critical section temperature T disqualification (T > b) is:
Figure GDA0003853552770000095
Figure GDA0003853552770000101
according to the basic properties of conditional probabilities, there are:
Figure GDA0003853552770000102
substituting the numerical value in the conditional probability table obtained in the step 2 to solve P r (x 1 ＝y 1 ,x 2 ＝y 2 ,…,x N ＝y N P | _ a), the probability that the critical section temperature T is qualified (T ≦ b) is finally calculated as P r (T≤b|x 1 ＝y 1 ,x 2 ＝y 2 ,…,x N ＝y N ,P⊙a)＝n％；
If n% <50%, judging that the critical section temperature T of the turboshaft engine to be predicted is unqualified under a Bayes classifier chain performance prediction model taking the target variable power P as a chain head; if n% is more than or equal to 50%, judging that the critical section temperature T of the turboshaft engine to be predicted is qualified under a Bayes classifier chain performance prediction model taking the target variable power P as a chain head;
and 4, step 4: the performance prediction results of the turboshaft engine on the target variable power P and the key section temperature T under a Bayes classifier chain performance prediction model based on a naive Bayes classifier and taking the target variable key section temperature T as a chain head are obtained through posterior probability reasoning; the specific process is as follows:
step 4-1: the Bayes classifier chain performance prediction model taking the target variable key section temperature T as the chain head comprises two classifiers, wherein the first classifier is used for predicting the target variable key section temperature T, the target variable key section temperature T is used as a father node, and each attribute variable x is subjected to prediction 1 ,x 2 ,…,x N As child nodes; the second classifier is used for predicting the target variable power P, taking the target variable power P as a parent node and classifying each attribute variable x 1 ,x 2 ,…,x N And the target variable key section temperature T is used as a child node;
step 4-2: predicting the probability of qualified critical section temperature T (T is less than or equal to b) according to a first classifier of a Bayes classifier chain performance prediction model taking the target variable critical section temperature T as a chain head;
the posterior probability of the key section temperature T qualified (T is less than or equal to b) is as follows:
Figure GDA0003853552770000103
the posterior probability of the critical section temperature T disqualification (T > b) is:
Figure GDA0003853552770000111
according to the basic properties of conditional probabilities, there are:
Figure GDA0003853552770000112
substituting the numerical value in the conditional probability table obtained in the step 2 to solve P r (x 1 ＝y 1 ,x 2 ＝y 2 ,…,x N ＝y N ) Finally, the probability that the critical section temperature T is qualified (T is less than or equal to b) is calculated and obtained as P r (T≤b|x 1 ＝y 1 ,x 2 ＝y 2 ,…,x N ＝y N )＝s％；
If s% <50%, judging that the critical section temperature T of the turboshaft engine to be predicted is unqualified under a Bayes classifier chain performance prediction model taking the target variable critical section temperature T as a chain head; if s% is more than or equal to 50%, judging that the key section temperature T of the turboshaft engine to be predicted is qualified under a Bayes classifier chain performance prediction model taking the key section temperature T of the target variable as a chain head;
step 4-3: predicting the probability of power P qualification (P > a) according to a second classifier of a Bayes classifier chain performance prediction model taking the target variable key section temperature T as a chain head;
the a posteriori probability of power phir (P > a) is then:
Figure GDA0003853552770000113
wherein, when s%<When the content of the organic acid is 50 percent,
Figure GDA0003853552770000114
is composed of>Number; when the s% is more than or equal to 50%,
Figure GDA0003853552770000115
is no more than number;
the posterior probability of unqualified power P (P ≦ a) is:
Figure GDA0003853552770000116
according to the basic properties of conditional probabilities, there are:
Figure GDA0003853552770000121
substituting the numerical value in the conditional probability table obtained in step 2 to solve
Figure GDA0003853552770000122
Figure GDA0003853552770000123
Finally, the power P qualified is obtained through calculation (P)>a) Has a probability of
Figure GDA0003853552770000124
If the r% is less than 50%, judging that the power P of the turboshaft engine to be predicted is unqualified under a Bayes classifier chain performance prediction model taking the target variable key section temperature T as a chain head; if the r% is more than or equal to 50%, judging that the power P of the turboshaft engine to be predicted is qualified under a Bayes classifier chain performance prediction model taking the key section temperature T of the target variable as a chain head;
and 5: averaging the qualification probability prediction results of the power P and the key section temperature T obtained in the step 3 and the step 4 under a Bayes classifier chain performance prediction model with the target variable power P and the target variable key section temperature T as chain heads respectively to finally obtain: the final qualification probability prediction result of the power P of the turboshaft engine to be predicted is
Figure GDA0003853552770000125
The final qualification probability prediction result of the critical section temperature T is
Figure GDA0003853552770000126
Step 6: if it is not
Figure GDA0003853552770000127
Finally judging that the power P of the turboshaft engine to be predicted is unqualified; if it is not
Figure GDA0003853552770000128
Finally judging that the power P of the turboshaft engine to be predicted is qualified;
if it is not
Figure GDA0003853552770000129
Finally judging that the critical section temperature T of the turboshaft engine to be predicted is unqualified; if it is used
Figure GDA00038535527700001210
And finally judging that the critical section temperature T of the turboshaft engine to be predicted is qualified.
The specific embodiment is as follows:
1. performance parameter data and design parameter data of the turboshaft engine are collected. The specific mode is as follows:
in the embodiment, the actual test run data of a batch of turboshaft engines of a certain model provided by a certain manufacturer is taken as a research object, and a plurality of main design parameters which are found in the long-term production process and have the greatest influence on the performance of the engine, namely three part size variables in the engine are respectively marked as X, Y and Z according to the suggestions of technicians of the manufacturer; and two performance parameter indexes, namely power P and the measurement data of the key section temperature T during test run. All data are continuous data.
2. And (3) determining a target variable and an attribute variable according to the data collected in the step (1), and setting a qualified condition of the target variable. Discretizing each variable to obtain a probability distribution table and a conditional probability table of two target variables and a conditional probability table of each state of the attribute variables of different target variables. The specific mode is as follows:
and (3) setting the power P and the critical section temperature T in the turboshaft engine data collected in the step (1) as target variables. According to the knowledge from manufacturers, on the premise of fully considering dynamic performance and safety, the minimum qualified power of a certain model of turboshaft engine is 1130KW, the temperature of the key section of the engine is not higher than 895 ℃ when the turboshaft engine is used, so that the qualified conditions of two target variables P and T are respectively 1130KW and 895 ℃, the two target variables are respectively discretized into two intervals by taking the qualified conditions as boundaries, and a probability distribution table and a conditional probability table are calculated. The method comprises the steps of setting size variables X, Y and Z of parts in three engines as attribute variables, respectively discretizing the attribute variables into three sections by adopting an equal frequency method, and respectively representing the three sections as states 0,1 and 2 for convenient representation, wherein 0,1 and 2 represent variable states and do not represent actual values. And counting the conditional probability tables of the three attribute variables relative to the two target variables.
TABLE 1 probability distribution Table for target variable P
Figure GDA0003853552770000131
TABLE 2 probability distribution Table for target variable T
Figure GDA0003853552770000132
TABLE 3 conditional probability table of target variables P
Figure GDA0003853552770000133
Table 4 conditional probability table of target variable T
Figure GDA0003853552770000134
Table 5 conditional probability table of attribute variables X with respect to P
Figure GDA0003853552770000141
Table 6 conditional probability table of attribute variables Y with respect to P
Figure GDA0003853552770000142
Table 7 conditional probability table of attribute variables Z with respect to P
Figure GDA0003853552770000143
Table 8 conditional probability table of attribute variables X with respect to T
Figure GDA0003853552770000144
Table 9 conditional probability table of attribute variables Y with respect to T
Figure GDA0003853552770000145
Table 10 conditional probability table of attribute variables Z with respect to T
Figure GDA0003853552770000146
3. And constructing a Bayesian classifier chain performance prediction model taking the target variable P as a chain head. The concrete method is as follows:
a. first, a first classifier in the Bayesian classifier chain performance prediction model is established. The first classifier is a prediction for the target variable P and can be obtained based on a naive bayes classifier based on the probability distribution table of the target variable P in table 1 obtained in step 2 and the conditional probability tables of the attribute variables with respect to the target variable P in tables 5 to 7. The graph of the classifier can be represented as pointing to three attribute variables from a target variable, wherein the target variable P is a parent node, and the attribute variables X, Y and Z are child nodes.
b. A second classifier in the Bayesian classifier chain performance prediction model is then established. The second classifier is for the prediction of the target variable T and can be based on a naive bayes classifier, and is obtained from the table 2 probability distribution table of the target variable T, the table 3 conditional probability table of the target variable P and the table 8-table 10 conditional probability tables of the attribute variables with respect to the target variable T, which are obtained in step 2. It should be noted that, in this classifier, the target variable T is the parent node, and the attribute variables X, Y, Z plus the target variable P of the first classifier are collectively used as the child node.
The finally established Bayesian classifier chain performance prediction model taking the target variable P as the chain head is shown in FIG. 2.
4. And (3) according to the Bayes classifier chain performance prediction model which is constructed in the step 3 and takes the target variable P as the chain head, combining a plurality of probability tables in the step 2, and for a specific turboshaft engine, obtaining the performance prediction results of the two target variables under the performance prediction model through posterior probability reasoning. The specific mode is as follows:
a. firstly, the performance of the target variable P is predicted according to a first classifier in a classifier chain performance prediction model. Assuming that the state values of the attribute variable combination (X, Y, Z) of the turboshaft engine to be predicted at present are (0, 1, 2), according to the naive Bayes characteristic of the first classifier constructed in the step 3, under the condition that the state of the target variable P is known, the state values of the attribute variables are mutually independent, and the posterior probability of P qualification (P > 1130) is as follows:
Figure GDA0003853552770000151
in the above formula, P in the molecule r (P>1130 64.03% from Table 1, P r (X＝0|P>1130)，P r (Y＝1|P>1130 ) and P r (Z＝2|P>1130 30.82%,30.14% and 28.77% from tables 5 to 7, respectively. These found values are substituted into the above equation:
Figure GDA0003853552770000152
the posterior probability of the same obtained P disqualification (P < = 1130) is as follows:
Figure GDA0003853552770000161
according to the basic properties of conditional probabilities, there are:
Figure GDA0003853552770000162
obtaining by solution:
P r (X＝0,Y＝1,Z＝2)＝0.04107053
obtaining:
Figure GDA0003853552770000163
in the bayesian classifier chain performance prediction model with the target variable P as the chain head, the qualification probability of the target variable P of the engine to be predicted is 41.66%, and the probability value is less than 50%, so that the engine is more likely to be in a disqualification state in view of the prediction result.
b. Next, the performance of the target variable T is predicted according to a second classifier in the classifier chain performance prediction model. In this classifier, there are four input variables: and X, Y, Z and P, wherein the values of X, Y and Z are known conditions (0, 1, 2), and the value of the state of P is set to be (P < = 1130) according to the prediction result of the first classifier. According to the naive Bayes characteristic of the second classifier constructed in the step 3, under the condition that the state of the target variable T is known, the state values of X, Y, Z and P are mutually independent, and the posterior probability of T qualification (T < = 895) is as follows:
Figure GDA0003853552770000164
p in the molecule of the above formula r (T. Ltoreq.895) is 67.54% in terms of P as shown in Table 2 r (X＝0|T≤895)，P r (Y＝1|T≤895)，P r (Z =2 calucin T. Ltoreq.895) and P r (P.ltoreq.1130 Y.ltoreq.895) 33.12%,37.66%,35.71% and 48.70% as seen in tables 8, 9, 10 and 3, respectively. These found values are substituted into the above equation:
Figure GDA0003853552770000171
the posterior probability of T disqualification (T > 895) is given by the same equation:
Figure GDA0003853552770000172
according to the basic properties of conditional probabilities, there are:
Figure GDA0003853552770000173
obtaining by solution:
P r (X＝0,Y＝1,Z＝2,P≤1130)＝0.01544833
obtaining:
Figure GDA0003853552770000174
in the Bayes classifier chain performance prediction model with the target variable P as the chain head, the qualification probability of the target variable T of the engine to be predicted is 94.84%.
In summary, in the bayesian classifier chain performance prediction model with the target variable P as the chain head, when the state value of the attribute variable combination (X, Y, Z) of the turboshaft engine to be predicted is (0, 1, 2), the qualification probability prediction result of the target variable P is 41.66%, and the state is biased to the unqualified state; the prediction result of the qualification probability of the target variable T is 94.84%, and the target variable T is biased to a qualification state.
5. And repeating the step 3 and the step 4, similarly constructing a Bayes classifier chain performance prediction model taking the target variable T as a chain head, and solving the performance prediction results of the two target variables under the performance prediction model for the same turboshaft engine to be predicted. The specific mode is as follows:
firstly, the step 3 is repeated, the sequence of P and T in the classifier chain is changed, and a Bayesian classifier chain performance prediction model which is established by adopting the same method and takes the target variable T as the chain head is shown in FIG. 3. The first classifier is used for predicting a target variable T, the target variable T is a father node, and attribute variables X, Y and Z are used as child nodes; the second classifier is used for predicting a target variable P, the target variable P is a father node, and the attribute variables X, Y and Z and the target variable T of the first classifier are taken as child nodes in a unified mode.
And then, by repeating the posterior probability reasoning process of the 4 th step, the performance prediction of two target variables is carried out aiming at the same turboshaft engine to be tested, and the following results can be obtained: in a Bayes classifier chain performance prediction model taking a target variable T as a chain head, when the state value of an attribute variable combination (X, Y, Z) of a turboshaft engine to be predicted is (0, 1, 2), the qualification probability prediction result of the target variable P is 29.70%, and the target variable P is biased to be in an unqualified state; the prediction result of the qualification probability of the target variable T was 78.11%, and the state was biased toward the qualification state.
6. And averaging the results obtained by the performance prediction models of the different classifier chains constructed in the 4 th step and the 5 th step to obtain the final performance prediction results of the two target variables. The concrete method is as follows:
the performance prediction model constructed by comprehensively considering two classifier chains in different sequences can be obtained by averaging the qualification probability prediction results of two target variables: when the state value of the attribute variable combination (X, Y, Z) of the turboshaft engine to be predicted is (0, 1, 2), the final qualified probability prediction result of the target variable P is 35.68%, and the final qualified probability prediction result is biased to be in an unqualified state; the final qualification probability prediction result of the target variable T is 86.48%, and the final qualification probability prediction result is biased to be in a qualification state. For any given turboshaft engine to be predicted, specific performance prediction results of two target variables of the turboshaft engine can be accurately and efficiently obtained through the method.

Claims (2)

1. A multi-target performance prediction method of a turboshaft engine based on a Bayesian classifier chain is characterized by comprising the following steps:
step 1: collecting performance parameter data and design parameter data of the turboshaft engine;
acquiring performance parameter data of the turboshaft engine by testing a plurality of turboshaft engines, wherein the performance parameter data comprises measurement data of power P and key section temperature T during test of the turboshaft engine;
obtaining design parameter data of a turboshaft engine, wherein the design parameter data comprises the sizes of N different parts in the turboshaft engine;
step 2: determining a target variable and an attribute variable, and setting a qualified condition of the target variable; discretizing each variable to obtain a probability distribution table and a conditional probability table of the target variable and conditional probability tables of various states of the attribute variables of different target variables; the method comprises the following specific steps:
step 2-1: setting the power P and the critical section temperature T of the turboshaft engine as target variables, and setting the sizes of N different parts in the turboshaft engine as attribute variables, wherein the x is represented as 1 ,x 2 ,…,x i ,…,x N ，i＝1,2,…,N；
Step 2-2: setting the minimum qualified condition of the power P as a KW; setting the maximum qualified condition of the temperature T of the key section as b ℃, wherein a and b respectively represent two normal constants;
step 2-3: dividing the power P into two sections by taking the minimum qualified condition of the power P as a boundary, and counting the probability distribution table P of the power P in the two sections by the data obtained in the step 1 r (P); dividing the critical section temperature T into two sections by taking the maximum qualified condition of the critical section temperature T as a boundary, and counting the probability distribution table P of the critical section temperature T in the two sections according to the data obtained in the step 1 r (T)；
The data statistical power P and the key section temperature T acquired in the step 1 are mutually prior conditional probability tables P belonging to different sections r (P | T) and P r (T|P)；
Step 2-4: each attribute variable x 1 ,x 2 ,…,x N Respectively dividing the obtained product into e segments by using an equal frequency method, and respectively representing the e segments in intervalsState 0,1, \8230, e-1,0,1, \8230, e-1 all represent attribute variable states and do not represent actual values of attribute variables; for the ith attribute variable, the threshold values adopted for dividing the attribute variable into e segments are respectively
Figure FDA0003853552760000011
Figure FDA0003853552760000012
In correspondence with the state 0, the state,
Figure FDA0003853552760000013
in correspondence with the state 1, the state of the mobile terminal,
Figure FDA0003853552760000014
corresponding to the state 2, and so on,
Figure FDA0003853552760000015
corresponding to state e-1;
step 2-5: counting N attribute variables by the data obtained in the step 1, and taking the power P and the key section temperature T as prior numbers to belong to a conditional probability table P of different states r (x i P) and P r (x i |T)；
And 3, step 3: obtaining a performance prediction result of the turboshaft engine on the target variable power P and the key section temperature T under a Bayes classifier chain performance prediction model based on a naive Bayes classifier and taking the target variable power P as a chain head through posterior probability reasoning; the specific process is as follows:
step 3-1: the Bayes classifier chain performance prediction model taking the target variable power P as the chain head comprises two classifiers, wherein the first classifier is used for predicting the target variable power P, the target variable power P is used as a father node, and each attribute variable x is used 1 ,x 2 ,…,x N As child nodes; the second classifier is used for predicting the key section temperature T of the target variable, taking the key section temperature T of the target variable as a father node and taking each attribute variable x 1 ,x 2 ,…,x N And the target variable power P as child nodes;
step 3-2: setting an attribute variable x 1 ,x 2 ,…,x N Respectively, is y 1 ,y 2 ,…,y N ，y 1 ,y 2 ,…,y N ∈[0,1,…,e-1](ii) a Predicting power Pqualified (P) according to a first classifier of a Bayes classifier chain performance prediction model taking target variable power P as a chain head>a probability of;
the posterior probability of power P qualification is then:
Figure FDA0003853552760000021
the posterior probability that the power P is unqualified, namely that P is less than or equal to a, is as follows:
Figure FDA0003853552760000022
according to the basic properties of conditional probabilities, there are:
Figure FDA0003853552760000023
substituting the numerical value in the conditional probability table obtained in step 2 to solve P r (x 1 ＝y 1 ,x 2 ＝y 2 ,…,x N ＝y N ) Finally, the probability of qualified power P is calculated to be P r (P>a|x 1 ＝y 1 ,x 2 ＝y 2 ,…,x N ＝y N )＝m％；
If m% is less than 50%, judging that the power P of the turboshaft engine to be predicted is unqualified under a Bayes classifier chain performance prediction model taking the target variable power P as a chain head; if m% is more than or equal to 50%, judging that the power P of the turboshaft engine to be predicted is qualified under a Bayes classifier chain performance prediction model taking the target variable power P as a chain head;
step 3-3: predicting the probability that the critical section temperature T is qualified, namely T is less than or equal to b according to a second classifier of a Bayes classifier chain performance prediction model taking the target variable power P as a chain head;
the posterior probability of the key section temperature T being qualified is:
Figure FDA0003853552760000031
wherein, when m% <50%, it is [ < no ]; when m% is more than or equal to 50%, the signal is more than or equal to the mark;
the posterior probability of the critical section temperature T disqualification, namely T > b is as follows:
Figure FDA0003853552760000032
according to the basic properties of conditional probabilities, there are:
Figure FDA0003853552760000033
substituting the numerical value in the conditional probability table obtained in the step 2 to solve P r (x 1 ＝y 1 ,x 2 ＝y 2 ,…,x N ＝y N P ^ a), the probability of obtaining the qualified critical section temperature T by final calculation is P r (T≤b|x 1 ＝y 1 ,x 2 ＝y 2 ,…,x N ＝y N ,P⊙a)＝n％；
If n% <50%, judging that the critical section temperature T of the turboshaft engine to be predicted is unqualified under a Bayes classifier chain performance prediction model taking the target variable power P as a chain head; if n% is more than or equal to 50%, judging that the critical section temperature T of the turboshaft engine to be predicted is qualified under a Bayes classifier chain performance prediction model taking the target variable power P as a chain head;
and 4, step 4: obtaining a performance prediction result of the turboshaft engine on the target variable power P and the key section temperature T under a Bayes classifier chain performance prediction model based on a naive Bayes classifier and taking the target variable key section temperature T as a chain head through posterior probability reasoning; the specific process is as follows:
step 4-1: the Bayes classifier chain performance prediction model taking the target variable key section temperature T as the chain head comprises two classifiers, wherein the first classifier is used for predicting the target variable key section temperature T, the target variable key section temperature T is used as a father node, and each attribute variable x is subjected to prediction 1 ,x 2 ,…,x N As child nodes; the second classifier takes the target variable power P as a parent node and takes each attribute variable x as the prediction of the target variable power P 1 ,x 2 ,…,x N And the target variable key section temperature T is taken as a child node;
step 4-2: predicting the qualified probability of the key section temperature T according to a first classifier of a Bayes classifier chain performance prediction model taking the target variable key section temperature T as a chain head;
the posterior probability of the key section temperature T being qualified is:
Figure FDA0003853552760000041
the posterior probability that the critical section temperature T is unqualified is as follows:
Figure FDA0003853552760000042
according to the basic properties of conditional probabilities, there are:
Figure FDA0003853552760000043
substituting the numerical value in the conditional probability table obtained in the step 2 to solve P r (x 1 ＝y 1 ,x 2 ＝y 2 ,…,x N ＝x N ) Finally, the probability of obtaining the qualification of the temperature T of the key section by calculation is P r (T≤b|x 1 ＝y 1 ,x 2 ＝y 2 ,…,x N ＝y N )＝s％；
If s% is less than 50%, judging that the critical section temperature T of the turboshaft engine to be predicted is unqualified under a Bayes classifier chain performance prediction model taking the target variable critical section temperature T as a chain head; if s% is more than or equal to 50%, judging that the key section temperature T of the turboshaft engine to be predicted is qualified under a Bayes classifier chain performance prediction model taking the key section temperature T of the target variable as a chain head;
step 4-3: predicting the qualified probability of the power P according to a second classifier of a Bayes classifier chain performance prediction model taking the target variable key section temperature T as a chain head;
the posterior probability of power qualification P is:
Figure FDA0003853552760000044
Figure FDA0003853552760000051
wherein, when s%<When the content of the organic acid is 50 percent,
Figure FDA00038535527600000510
is composed of>Number; when the s% is more than or equal to 50%,
Figure FDA00038535527600000511
is no more than number;
the posterior probability that the power P is unqualified is as follows:
Figure FDA0003853552760000052
according to the basic properties of conditional probabilities, there are:
Figure FDA0003853552760000053
substituting the numerical value in the conditional probability table obtained in step 2 to solve
Figure FDA00038535527600000512
Figure FDA00038535527600000513
The probability of qualified power P is finally calculated to be
Figure FDA00038535527600000514
If the r% is less than 50%, judging that the power P of the turboshaft engine to be predicted is unqualified under a Bayes classifier chain performance prediction model taking the target variable key section temperature T as a chain head; if the r% is more than or equal to 50%, judging that the power P of the turboshaft engine to be predicted is qualified under a Bayes classifier chain performance prediction model taking the key section temperature T of the target variable as a chain head;
and 5: averaging the qualification probability prediction results of the power P and the key section temperature T obtained in the step 3 and the step 4 under a Bayes classifier chain performance prediction model taking the target variable power P and the target variable key section temperature T as chain heads respectively to finally obtain: the final qualification probability prediction result of the power P of the turboshaft engine to be predicted is
Figure FDA0003853552760000054
The final qualification probability prediction result of the critical section temperature T is
Figure FDA0003853552760000055
Step 6: if it is not
Figure FDA0003853552760000056
Finally judging that the power P of the turboshaft engine to be predicted is unqualified; if it is not
Figure FDA0003853552760000057
Finally judging that the power P of the turboshaft engine to be predicted is qualified;
if it is used
Figure FDA0003853552760000058
Finally judging that the critical section temperature T of the turboshaft engine to be predicted is unqualified; if it is not
Figure FDA0003853552760000059
And finally judging that the critical section temperature T of the turboshaft engine to be predicted is qualified.
2. The turbo shaft engine multi-target performance prediction method based on the Bayesian classifier chain as recited in claim 1, wherein e =3.
CN202110140394.0A 2021-02-02 2021-02-02 Turboshaft engine multi-target performance prediction method based on Bayesian classifier chain Active CN112765899B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110140394.0A CN112765899B (en) 2021-02-02 2021-02-02 Turboshaft engine multi-target performance prediction method based on Bayesian classifier chain

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110140394.0A CN112765899B (en) 2021-02-02 2021-02-02 Turboshaft engine multi-target performance prediction method based on Bayesian classifier chain

Publications (2)

Publication Number Publication Date
CN112765899A CN112765899A (en) 2021-05-07
CN112765899B true CN112765899B (en) 2022-11-25

Family

ID=75704589

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110140394.0A Active CN112765899B (en) 2021-02-02 2021-02-02 Turboshaft engine multi-target performance prediction method based on Bayesian classifier chain

Country Status (1)

Country Link
CN (1) CN112765899B (en)

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111651940A (en) * 2020-06-11 2020-09-11 中国人民解放军海军航空大学 Turboshaft engine steady-state model identification method based on PSO-NARX
CN111814272A (en) * 2020-07-07 2020-10-23 中国科学院工程热物理研究所 Turbine pneumatic-dynamic response intelligent optimization design method based on machine learning

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8977575B2 (en) * 2010-07-20 2015-03-10 Lockheed Martin Corporation Confidence level generator for bayesian network
CN102411305B (en) * 2011-11-11 2014-03-05 南京航空航天大学 Design method of comprehensive disturbance rejection control system for single-rotor wing helicopter/turboshaft engine
US11124317B2 (en) * 2018-01-26 2021-09-21 Dalian University Of Technology Method for prediction of key performance parameters of aero-engine in transition condition
CN110826587B (en) * 2019-08-15 2024-03-19 南京航空航天大学 Vortex shaft engine fault detection method based on improved weighted support vector machine
CN111814350A (en) * 2020-07-22 2020-10-23 西北工业大学 Turbine engine performance optimization method based on Bayesian network and importance

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111651940A (en) * 2020-06-11 2020-09-11 中国人民解放军海军航空大学 Turboshaft engine steady-state model identification method based on PSO-NARX
CN111814272A (en) * 2020-07-07 2020-10-23 中国科学院工程热物理研究所 Turbine pneumatic-dynamic response intelligent optimization design method based on machine learning

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
基于神经网络的涡轴发动机故障诊断仿真研究;刘岩松等;《内燃机与配件》;20200115(第01期);第6-7页 *

Also Published As

Publication number Publication date
CN112765899A (en) 2021-05-07

Similar Documents

Publication Publication Date Title
CN109931678B (en) Air conditioner fault diagnosis method based on deep learning LSTM
CN107992976B (en) Hot topic early development trend prediction system and prediction method
CN111062508B (en) Method for evaluating real-time running state of wind turbine generator based on big data technology
CN110571792A (en) Analysis and evaluation method and system for operation state of power grid regulation and control system
CN111638707B (en) Intermittent process fault monitoring method based on SOM clustering and MPCA
CN112749840B (en) Method for acquiring energy efficiency characteristic index reference value of thermal power generating unit
CN113313139A (en) Wind power prediction uncertainty quantification method based on dynamic characteristics of unit
CN116805785B (en) Power load hierarchy time sequence prediction method based on random clustering
CN111814350A (en) Turbine engine performance optimization method based on Bayesian network and importance
CN113408659A (en) Building energy consumption integrated analysis method based on data mining
CN107357994B (en) Staged mining method for aircraft engine performance decline mode
Huang et al. Bayesian neural network based method of remaining useful life prediction and uncertainty quantification for aircraft engine
CN112381965A (en) Aeroengine health state identification system and method based on data mining
CN112765899B (en) Turboshaft engine multi-target performance prediction method based on Bayesian classifier chain
CN111126477A (en) Learning and reasoning method of hybrid Bayesian network
Rasay et al. Reinforcement learning based on stochastic dynamic programming for condition-based maintenance of deteriorating production processes
CN114384423A (en) Fuel cell health state identification method based on typical operation parameters
CN117633688A (en) Large-scale power data anomaly detection method based on ridge regression-k-means clustering-LOF-LSTM fusion algorithm
CN115907079B (en) Airspace traffic flow prediction method based on attention space-time diagram convolutional network
Hatherall et al. Load prediction based remaining discharge energy estimation using a combined online and offline prediction framework
Ma et al. State-of-Health assessment for aero-engine based on density-distance clustering and fuzzy Bayesian risk
Chen et al. A data fusion-based methodology of constructing health indicators for anomaly detection and prognostics
CN115718880A (en) Prediction method for degradation stage of complex equipment
Kozlenko et al. Deep learning based fault detection of natural gas pumping unit
Wang et al. Degradation pattern identification and remaining useful life prediction for mechanical equipment using SKF-EN

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