CN112418306A - Gas turbine compressor fault early warning method based on LSTM-SVM - Google Patents
Gas turbine compressor fault early warning method based on LSTM-SVM Download PDFInfo
- Publication number
- CN112418306A CN112418306A CN202011309245.4A CN202011309245A CN112418306A CN 112418306 A CN112418306 A CN 112418306A CN 202011309245 A CN202011309245 A CN 202011309245A CN 112418306 A CN112418306 A CN 112418306A
- Authority
- CN
- China
- Prior art keywords
- compressor
- fault
- early warning
- svm
- alarm
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 30
- 208000024891 symptom Diseases 0.000 claims abstract description 26
- 238000012544 monitoring process Methods 0.000 claims abstract description 15
- 238000012549 training Methods 0.000 claims abstract description 12
- 238000012545 processing Methods 0.000 claims abstract description 8
- 238000013135 deep learning Methods 0.000 claims abstract description 4
- 239000013598 vector Substances 0.000 claims description 15
- 238000013145 classification model Methods 0.000 claims description 13
- 238000012360 testing method Methods 0.000 claims description 5
- 238000012706 support-vector machine Methods 0.000 claims description 3
- 230000006870 function Effects 0.000 claims description 2
- 238000011160 research Methods 0.000 description 7
- 230000008859 change Effects 0.000 description 4
- 238000005299 abrasion Methods 0.000 description 3
- 238000005260 corrosion Methods 0.000 description 3
- 230000007797 corrosion Effects 0.000 description 3
- 230000007423 decrease Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000010248 power generation Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000009825 accumulation Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 239000000428 dust Substances 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003746 surface roughness Effects 0.000 description 1
Images
Classifications
-
- 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
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Development Economics (AREA)
- Evolutionary Biology (AREA)
- Game Theory and Decision Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Entrepreneurship & Innovation (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Control Of Positive-Displacement Air Blowers (AREA)
Abstract
The invention relates to a gas turbine compressor fault early warning method based on an LSTM-SVM, which comprises the following steps: establishing a fault knowledge base of the gas compressor of the gas turbine, and excavating the relation between the fault type and the fault symptom of the gas compressor. Determining the symptom parameter type of the compressor; training the normal data of the compressor symptom parameters by using a deep learning LSTM algorithm to establish a good prediction model; monitoring a data curve output by the model, performing data processing on alarm data through positive and negative deviation degrees, performing fault classification as an input parameter of the SVM, and determining the fault type of the gas compressor. According to the early warning method and the early warning system, the fault trend of the gas compressor can be quickly found through early warning information, and important decision support is provided for early warning of the fault of the gas compressor.
Description
Technical Field
The invention relates to the field of thermal power plant equipment fault early warning, in particular to a gas turbine compressor fault early warning method based on an LSTM-SVM.
Background
In the power industry, a gas turbine generator set becomes one of the mainstream power generation modes at present due to the advantages of quick start and stop, high heat efficiency, less pollution and the like, but the key core technology of the gas turbine in China still depends on foreign countries. In order to change the current situation, the state increases the policy support for the gas turbine, and promotes the rapid development of the gas turbine industry. The compressor is one of the important parts of the gas turbine, and the running state of the compressor directly influences the safety and the reliability of the gas turbine. However, when the compressor operates in a high-speed and high-temperature environment for a long time, faults such as blade fouling, abrasion corrosion and the like often occur. If the fault trend of the gas turbine compressor can be found early, the faults are repaired and protected in advance, and the risk of unstable operation or unplanned shutdown of the gas turbine caused by the fault of the compressor is reduced. Therefore, the early warning of the faults of the gas compressor has important significance on the stable operation of the gas turbine.
At present, the research on the air compressor mainly focuses on the influence of the type of the air compressor fault on the performance of the air turbine unit, the research on the early warning of the air compressor fault is less, and the method and the achievement for the early warning of the power plant equipment fault are provided. The method comprises the steps of carrying out fault early warning on a power plant fan by utilizing a multivariate state estimation and deviation degree method, establishing a parameter model of the fan through multivariate state estimation, outputting a result by utilizing a deviation degree monitoring model, capturing a fault development process and realizing early warning. The power generation equipment fault early warning system based on the similarity principle establishes a model matrix by mathematical analysis of historical data, compares an actual value with a model output estimation value, and gives an alarm when a preset deviation is exceeded. The dynamic early warning model of the steam turbine generator set is built by using a grey theory and a similarity principle, and the abnormal state of the equipment can be found in time by using a hypersphere similarity analysis technology, so that a new method is provided for early warning of equipment faults. Although the method is not applied to the field of gas turbines, the method has certain reference value, the structure of the gas compressor is complex, the change relevance of corresponding characteristic parameters among various fault types is strong, and the early fault early warning difficulty of the gas turbine gas compressor is increased.
Disclosure of Invention
The invention aims to solve the early fault early warning problem of the gas turbine compressor.
In order to achieve the aim, the technical scheme of the invention provides a gas turbine compressor fault early warning method based on an LSTM-SVM, which is characterized in that an LSTM prediction model is established by using normal historical data of the compressor, and the output prediction error of the LSTM prediction model is within 0.5%. And then substituting 7 characteristic parameters of the air compressor air inlet flow, efficiency, inlet temperature, outlet temperature, inlet pressure, outlet pressure and stage pack pressure ratio into an LSTM prediction model to obtain a residual error curve of a predicted value and an actual value, setting an early warning threshold value, and alarming for overrun. And finally, performing data processing on alarm information of the alarm point through positive and negative deviation degrees, and performing fault classification as an input parameter of the SVM, so as to determine the fault type of the gas compressor and further realize early fault early warning of the gas turbine gas compressor, wherein the method specifically comprises the following steps:
s1, establishing a fault knowledge base of the gas turbine compressor, wherein the fault knowledge base comprises the relation between the type of the compressor fault and the compressor symptom parameters of the fault; simply and intuitively find the changes of the fault type of the gas compressor and the related gas compressor symptom parameters through a fault knowledge base, analyze and reason the changes, dig out the relation between the fault type of the gas compressor and the symptom parameters of the fault gas compressor, and determine which gas compressor symptom parameters have larger influence when the gas compressor has faults;
s2, determining the variety of the compressor symptom parameters, training the determined normal data of the compressor symptom parameters by using a deep learning LSTM algorithm, and establishing a prediction model to control the residual error range between the predicted value and the actual value output by the prediction model within 0.5%;
s3, substituting the actual value of the compressor symptom parameter into the prediction model established in the step S2 for monitoring, alarming when the residual error between the predicted value and the actual value output by the prediction model exceeds 0.9%, and carrying out positive and negative deviation degree processing on alarm information, wherein the alarm information is the data value of the compressor symptom parameter of an alarm point;
and S4, classifying the positive and negative deviation degrees of the alarm information as input parameters of the SVM classification model, determining the fault type of the compressor, and realizing fault early warning of the compressor.
Preferably, in the step S3, an alarm is given by monitoring a residual error curve of compressor characteristic parameters, where the compressor characteristic parameters are compressor intake air flow, efficiency, inlet temperature, outlet temperature, inlet pressure, outlet pressure, and stage group pressure ratio, which are respectively marked as x1,x2,…,x7；
Setting alarm limits which comprise a high alarm limit and a low alarm limit, alarming when residual errors between predicted values and actual values output by the prediction model exceed the limits, and uploading alarm information;
and (2) carrying out positive and negative deviation degree processing on the alarm information by using the following formula (1):
in formula (1), i is 1,2, …, 7; m isiThe deviation degree of an alarm point of the ith compressor characteristic parameter is obtained; x is the number ofi ShiThe actual value of the ith compressor characteristic parameter is obtained; x is the number ofi predictionThe predicted value of the ith compressor characteristic parameter is obtained;
defining the exceeding of the high alarm limit as positive deviation, namely the residual monitoring exceeds the upper limit; the over-low alarm limit is defined as negative deviation degree, namely residual error monitoring over-low limit; if no alarm information exists, the alarm information is defined as 0;
by LSTM preThe measuring model monitors 7 compressor characteristic parameters, so that the alarm point deviation degrees corresponding to the 7 compressor characteristic parameters form a vector group Mk，Mk＝{m1,m2,…,m7And k is the number of samples of the alarm information vector group and is used as the input of the SVM classification model, wherein the training sample k is 20, and the test sample k is 80.
Preferably, in step S4, N groups of alarm vectors are obtained from a residual curve output by the prediction model as an input of an SVM classification model, and an SVM is used for classification to determine a fault type of the compressor, where an SVM classification model training set T is shown as the following formula (2):
T＝{(M1,y1),(M2,y2),…,(Mk,yk),…,(MN,yN)} (2)
in the formula (2), MkThe k alarm deviation vector group is obtained; y iskFor compressor fault type marking, yk＝{-1，1}；
The objective function and constraint conditions are shown in the following equations (3) and (4):
s.t.yk(wmk+b)≥1,k＝1,2,...,N (4)
in the formulas (3) and (4), w is a hyperplane normal vector for dividing the fault type of the compressor, and b is a hyperplane offset term for dividing the fault type of the compressor.
Preferably, in step S4, if there are 4 fault categories in the gas turbine compressor fault early warning, in the SVM classification model, to solve the problem of classification of the support vector machine 4, a method for classifying the remaining categories is adopted, that is, a certain category is first classified, the other three categories are classified into another category to form a sample set, and after the four operations are completed, 4 sample sets are obtained, so that 4 binary classification problems are formed, each sample set has a classification result, and the maximum value is selected as the final classification result.
The invention provides a gas turbine compressor fault early warning method based on an LSTM-SVM, aiming at the early fault early warning problem of the gas turbine compressor. Establishing an LSTM prediction model through historical data of the gas compressor, substituting characteristic parameters of the gas compressor into the prediction model, monitoring a residual curve between a predicted value and an actual value output by the model, setting an alarm threshold value, alarming in an overrun mode, carrying out data processing on alarm information of an alarm point through positive and negative deviation degrees, using the alarm information as an input parameter of a support vector machine, and identifying the fault type of the gas compressor, thereby realizing the fault early warning of the gas turbine gas compressor.
The method can be used for rapidly and accurately finding the early failure trend of the gas turbine compressor, timely maintaining and protecting the gas turbine compressor, reducing the economic loss caused by the failure of the gas turbine compressor and ensuring that the gas turbine can reliably and safely operate. Compared with the prior art, the invention has the following specific advantages:
(1) the prediction model established by the deep learning LSTM algorithm has a good prediction effect on time samples, is very suitable for the compressor operation data sampled by time, and has small prediction error.
(2) The method has the advantages that a compressor fault knowledge base is constructed by utilizing research results of student documents and compressor fault case analysis, and the relation between the compressor fault type and the symptom parameters can be intuitively and simply reflected.
(3) The SVM is suitable for training of small samples, the classification speed is high, the final result is determined by a few support vectors, the SVM can help people to grasp key samples, and the SVM has good robustness.
Drawings
FIG. 1 is a diagram of a gas turbine compressor fault early warning method structure of the LSTM-SVM of the present invention;
FIG. 2 is an overall flow of gas turbine compressor fault warning;
FIG. 3 shows an actual value and a predicted value of the air intake amount of the compressor of the gas turbine;
FIG. 4 is a curve of predicted values and actual values of air inflow of the compressor;
FIG. 5 is a compressor air intake amount residual percentage;
FIG. 6 is a predicted value and an actual value of the air compressor blade fouling air inflow;
FIG. 7 is a compressor blade fouling intake air quantity residual error;
FIG. 8 shows predicted values and actual values of efficiency of a blade fouling compressor;
FIG. 9 is a blade fouling compressor efficiency residual;
FIG. 10 is a sample LSTM-SVM training sample diagnostic result;
FIG. 11 shows the results of the LSTM-SVM test.
Detailed Description
The invention will be further illustrated with reference to the following specific examples. It should be understood that these examples are for illustrative purposes only and are not intended to limit the scope of the present invention. Further, it should be understood that various changes or modifications of the present invention may be made by those skilled in the art after reading the teaching of the present invention, and such equivalents may fall within the scope of the present invention as defined in the appended claims.
As shown in figure 1, the fault early warning method for the gas turbine compressor integrally adopts an LSTM-SVM. And establishing a fault knowledge base of the gas turbine compressor according to expert knowledge, experience and the compressor fault case, wherein the knowledge base mainly comprises fault types and fault symptom parameters. The compressor fault knowledge base can simply and visually find the change relation between the compressor fault type and the related symptom parameters, so that 7 symptom parameters of the compressor air inlet flow, efficiency, inlet temperature, outlet temperature, inlet pressure, outlet pressure and stage pack pressure ratio are determined, then 7 characteristic parameters of the compressor air inlet flow, the efficiency … stage pack pressure ratio and the like are substituted into an LSTM prediction model, a residual error curve of a predicted value and an actual value is obtained, an early warning threshold value is set, and the alarm is carried out in an overrun mode. And finally, data processing is carried out on alarm information of the alarm point through positive and negative deviation degrees, the alarm information is used as an input parameter of the SVM to carry out fault classification, and the fault type of the gas compressor is determined, so that the fault early warning of the gas turbine gas compressor is realized. The specific flow chart is shown in fig. 2, and the fault early warning method specifically comprises the following steps:
s1, firstly, determining the research object. A gas turbine PG9351FA of a certain power plant is taken as a research object, the performance parameters of a compressor of the gas turbine PG9351FA are that the atmospheric pressure is 101.3kpa, the output power is 255.6MW, the pressure ratio of the compressor is 15.4, and the like. The compressor often breaks down in the environment of high temperature and high rotation speed after long-term operation, and the common faults of the compressor are compressor blade fouling, blade abrasion corrosion, air inlet icing and compressor surge. Compressor blade fouling is taken as an example. In the operation process of the compressor, dust in the air can be sucked into the compressor, and the surface roughness of a rotor of the compressor is increased along with long-time accumulation. Compressor fouling affects changes in compressor symptom parameters, thereby reducing the operating efficiency of the gas turbine. Therefore, the fault knowledge base of the gas compressor can be constructed to simply and intuitively discover the fault type of the gas compressor and the change relation of related symptom parameters of the gas compressor, and provide a basis for a rule reasoning fault diagnosis method. According to the research results of expert scholars and the fault case analysis of the compressor, a fault knowledge base of the compressor is constructed, as shown in fig. 3.
And S2, selecting research data. According to a compressor fault knowledge base, the hidden relation between the compressor fault type and the symptom parameter is excavated, and 7 characteristic parameters of the compressor air inlet flow, inlet temperature, outlet temperature, inlet pressure, outlet pressure, stage group pressure ratio and compressor efficiency are selected as indexes for reflecting the state of the compressor of the gas turbine and are respectively marked as x1,x2,…,x7. And monitoring the residual error curves of the predicted values and the actual values of the 7 characteristic parameters, capturing the early fault characteristics of the gas compressor, and providing a basis for fault identification and early warning of the gas turbine gas compressor. By training the normal air inflow of the gas compressor of the gas turbine and observing a curve of a predicted value and an actual value output by the LSTM prediction model, the LSTM prediction model has good goodness of fit, as shown in figure 4. Fig. 5 is a residual percentage curve of the predicted value and the actual value of the normal air intake quantity of the compressor output by the model, and as seen from fig. 5, the error of the residual percentage curve of the predicted value and the actual value of the air intake quantity of the compressor is within 0.5%, so that the prediction effect is good.
S3, in order to capture the fault trend of the gas turbine compressor, selecting early fault data of a segment of compressor blade accumulated dirt air inflow to be brought into an LSTM prediction model, and obtaining the predicted value and the accumulated dirt air inflow of the compressor bladeActual values, as shown in fig. 6, it can be seen from fig. 6 that the actual value of the intake air amount starts to shift downward when the compressor blades are fouled, and the compressor performance starts to decline. By monitoring a residual curve of an air compressor air inflow predicted value and an actual value, as shown in fig. 7, the fluctuation of the air compressor blade accumulated dirt at the beginning of the air inflow residual curve is uniform, an alarm is given when the air compressor blade accumulated dirt exceeds an alarm limit at a 116 th point, the actual value of the alarm point is 616.48kg/s, the predicted value is 623.08kg/s, and the negative deviation degree is-1.05 according to a deviation degree formula (1) and a low alarm. Compressor efficiency also tends to decline for compressor blade fouling failures, as shown in fig. 8. By monitoring a residual error curve of a predicted value and an actual value of the compressor efficiency of the blade fouling, as shown in fig. 9, the fluctuation of the residual error of the compressor blade fouling efficiency is normal at the beginning, and the residual error exceeds an alarm limit by 148 th point, so that an alarm is given. The actual value of the alarm point is 87.15%, and the predicted value is 87.92%. And obtaining the negative deviation degree of-0.87 according to the deviation degree formula (1) and the low alarm. Through the method, positive and negative deviation degrees of other characteristic parameters of the compressor blade scale are obtained in the same way to form an alarm vector group M1{ -1.05, -0.87,0, 0.93, 0, -0.97, -1.12}, which are used as input parameters of the SVM classification model for fault identification.
S4, aiming at a PG9351FA type gas turbine, the common faults of a compressor mainly comprise 4 types, namely blade fouling, blade abrasion corrosion, air inlet icing and surging, and the 4 fault types are respectively marked as 1,2, 3 and 4. And acquiring an alarm deviation vector group from the LSTM prediction model as the input of the SVM classification model for classification and identification. Firstly, 5 alarm vector groups are selected for each fault type to be trained, 20 training samples are provided in total, the training results are shown in fig. 10, and the accuracy of the classification results is high. Test samples are selected, 20 alarm vector groups are selected for each fault type to be tested, the total number of the test samples is 80, and SVM classification results are shown in FIG. 11. The accuracy rate of 98.7% can be seen from the SVM classification result in FIG. 11.
An LSTM prediction model is established for characteristic parameters of a gas compressor of the gas turbine, a residual error curve of predicted values and actual values of the characteristic parameters is monitored, the early failure trend of the gas compressor is captured through fluctuation of residual errors to give an alarm, and the alarm information of an alarm point is quantized by introducing positive and negative deviation degrees. And finally, judging the fault type of the gas turbine compressor by using the SVM classification model, and providing method reference for early fault early warning work of the gas turbine compressor.
Claims (4)
1. A gas turbine compressor fault early warning method based on an LSTM-SVM is characterized by comprising the following steps:
s1, establishing a fault knowledge base of the gas turbine compressor, wherein the fault knowledge base comprises the relation between the type of the compressor fault and the compressor symptom parameters of the fault; simply and intuitively find the changes of the fault type of the gas compressor and the related gas compressor symptom parameters through a fault knowledge base, analyze and reason the changes, dig out the relation between the fault type of the gas compressor and the symptom parameters of the fault gas compressor, and determine which gas compressor symptom parameters have larger influence when the gas compressor has faults;
s2, determining the variety of the compressor symptom parameters, training the determined normal data of the compressor symptom parameters by using a deep learning LSTM algorithm, and establishing a prediction model to control the residual error range between the predicted value and the actual value output by the prediction model within 0.5%;
s3, substituting the actual value of the compressor symptom parameter into the prediction model established in the step S2 for monitoring, alarming when the residual error between the predicted value and the actual value output by the prediction model exceeds 0.9%, and carrying out positive and negative deviation degree processing on alarm information, wherein the alarm information is the data value of the compressor symptom parameter of an alarm point;
and S4, classifying the positive and negative deviation degrees of the alarm information as input parameters of the SVM classification model, determining the fault type of the compressor, and realizing fault early warning of the compressor.
2. The LSTM-SVM-based gas turbine compressor failure early warning method as claimed in claim 1, wherein the step S3 is performed by monitoring a residual error curve of the compressor characteristic parameters, wherein the compressor characteristic parameters are compressor inlet air flow, efficiency, inlet temperature, outlet temperature, inlet temperature and outlet temperatureThe port pressure, outlet pressure, and stage group pressure ratio, each denoted as x1,x2,…,x7；
Setting alarm limits which comprise a high alarm limit and a low alarm limit, alarming when residual errors between predicted values and actual values output by the prediction model exceed the limits, and uploading alarm information;
and (2) carrying out positive and negative deviation degree processing on the alarm information by using the following formula (1):
in formula (1), i is 1,2, …, 7; m isiThe deviation degree of an alarm point of the ith compressor characteristic parameter is obtained; x is the number ofi ShiThe actual value of the ith compressor characteristic parameter is obtained; x is the number ofi predictionThe predicted value of the ith compressor characteristic parameter is obtained;
defining the exceeding of the high alarm limit as positive deviation, namely the residual monitoring exceeds the upper limit; the over-low alarm limit is defined as negative deviation degree, namely residual error monitoring over-low limit; if no alarm information exists, the alarm information is defined as 0;
monitoring 7 compressor characteristic parameters through an LSTM prediction model, and forming a vector group M by alarm point deviation degrees corresponding to the 7 compressor characteristic parametersk，Mk＝{m1,m2,…,m7And k is the number of samples of the alarm information vector group and is used as the input of the SVM classification model, wherein the training sample k is 20, and the test sample k is 80.
3. The LSTM-SVM-based gas turbine compressor fault early warning method as claimed in claim 1, wherein in step S4, N groups of alarm vectors are obtained from the residual curve outputted from the prediction model as the input of the SVM classification model, and classification is performed by using SVM to determine the fault type of the compressor, wherein the training set T of the SVM classification model is as shown in the following formula (2):
T＝{(M1,y1),(M2,y2),…,(Mk,yk),…,(MN,yN)} (2)
in the formula (2), MkThe k alarm deviation vector group is obtained; y iskFor compressor fault type marking, yk＝{-1，1}；
The objective function and constraint conditions are shown in the following equations (3) and (4):
s.t.yk(wmk+b)≥1,k＝1,2,...,N (4)
in the formulas (3) and (4), w is a hyperplane normal vector for dividing the fault type of the compressor, and b is a hyperplane offset term for dividing the fault type of the compressor.
4. The LSTM-SVM-based gas turbine compressor failure early warning method as claimed in claim 1, wherein in step S4, if there are 4 failure categories in the gas turbine compressor failure early warning, in the SVM classification model, for solving the problem of classification of the support vector machine 4, a method of classifying the remaining categories is adopted, that is, a certain category is first classified, the other three categories are classified into another category to form a sample set, after four times of the above operations, 4 sample sets are obtained, thereby forming 4 binary categories, each sample set has a classification result, and the maximum value is selected as the final classification result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011309245.4A CN112418306B (en) | 2020-11-20 | 2020-11-20 | Gas turbine compressor fault early warning method based on LSTM-SVM |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011309245.4A CN112418306B (en) | 2020-11-20 | 2020-11-20 | Gas turbine compressor fault early warning method based on LSTM-SVM |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112418306A true CN112418306A (en) | 2021-02-26 |
| CN112418306B CN112418306B (en) | 2024-03-29 |
Family
ID=74773811
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011309245.4A Active CN112418306B (en) | 2020-11-20 | 2020-11-20 | Gas turbine compressor fault early warning method based on LSTM-SVM |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112418306B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112632805A (en) * | 2021-03-15 | 2021-04-09 | 国能大渡河大数据服务有限公司 | Analysis early warning method, system, terminal and medium for crossing vibration area of unit |
| CN115130559A (en) * | 2022-06-06 | 2022-09-30 | 中国船舶集团有限公司***工程研究院 | Method, system and terminal for monitoring starting process and evaluating state of marine gas turbine |
| CN117516927A (en) * | 2024-01-05 | 2024-02-06 | 四川省机械研究设计院(集团)有限公司 | Gearbox fault detection method, system, equipment and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109814527A (en) * | 2019-01-11 | 2019-05-28 | 清华大学 | Based on LSTM Recognition with Recurrent Neural Network industrial equipment failure prediction method and device |
| CN110161343A (en) * | 2019-06-12 | 2019-08-23 | 中南大学 | A kind of non-intrusion type real-time dynamic monitoring method of intelligence train exterior power receiving device |
| US20190324439A1 (en) * | 2017-08-02 | 2019-10-24 | Strong Force Iot Portfolio 2016, Llc | Data monitoring systems and methods to update input channel routing in response to an alarm state |
| CN111237134A (en) * | 2020-01-14 | 2020-06-05 | 上海电力大学 | Offshore double-fed wind driven generator fault diagnosis method based on GRA-LSTM-stacking model |
| CN111340110A (en) * | 2020-02-25 | 2020-06-26 | 湖南师范大学 | Fault early warning method based on industrial process running state trend analysis |
-
2020
- 2020-11-20 CN CN202011309245.4A patent/CN112418306B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190324439A1 (en) * | 2017-08-02 | 2019-10-24 | Strong Force Iot Portfolio 2016, Llc | Data monitoring systems and methods to update input channel routing in response to an alarm state |
| CN109814527A (en) * | 2019-01-11 | 2019-05-28 | 清华大学 | Based on LSTM Recognition with Recurrent Neural Network industrial equipment failure prediction method and device |
| CN110161343A (en) * | 2019-06-12 | 2019-08-23 | 中南大学 | A kind of non-intrusion type real-time dynamic monitoring method of intelligence train exterior power receiving device |
| CN111237134A (en) * | 2020-01-14 | 2020-06-05 | 上海电力大学 | Offshore double-fed wind driven generator fault diagnosis method based on GRA-LSTM-stacking model |
| CN111340110A (en) * | 2020-02-25 | 2020-06-26 | 湖南师范大学 | Fault early warning method based on industrial process running state trend analysis |
Non-Patent Citations (2)
| Title |
|---|
| 何群;尹飞飞;武鑫;谢平;江国乾;: "基于长短期记忆网络的风电机组齿轮箱故障预测", 计量学报, no. 10, 26 October 2020 (2020-10-26) * |
| 黄佳林;茅大钧;: "基于多维时间序列状态估计的压气机故障预警模型建立", 上海电机学院学报, no. 04, 25 August 2017 (2017-08-25) * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112632805A (en) * | 2021-03-15 | 2021-04-09 | 国能大渡河大数据服务有限公司 | Analysis early warning method, system, terminal and medium for crossing vibration area of unit |
| CN115130559A (en) * | 2022-06-06 | 2022-09-30 | 中国船舶集团有限公司***工程研究院 | Method, system and terminal for monitoring starting process and evaluating state of marine gas turbine |
| CN117516927A (en) * | 2024-01-05 | 2024-02-06 | 四川省机械研究设计院(集团)有限公司 | Gearbox fault detection method, system, equipment and storage medium |
| CN117516927B (en) * | 2024-01-05 | 2024-04-05 | 四川省机械研究设计院(集团)有限公司 | Gearbox fault detection method, system, equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112418306B (en) | 2024-03-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112418306A (en) | Gas turbine compressor fault early warning method based on LSTM-SVM | |
| KR101955305B1 (en) | Gas turbine sensor failure detection utilizing a sparse coding methodology | |
| US6587737B2 (en) | Method for the monitoring of a plant | |
| CN111648992B (en) | Gas turbine compressor fault identification early warning method | |
| CN112067335A (en) | Power plant blower fault early warning method based on multivariate state estimation | |
| CN108022058B (en) | Wind turbine state reliability assessment method | |
| CN112257943A (en) | Power plant induced draft fan fault early warning method based on MSET and deviation degree | |
| CN110209144B (en) | Two-layer real-time monitoring and alarm tracing method based on dynamic and static cooperative difference analysis | |
| WO2016208315A1 (en) | Plant diagnosis device and plant diagnosis method | |
| CN114757380B (en) | Fault early warning system and method for thermal power plant, electronic equipment and storage medium | |
| CN114239708B (en) | Combustion engine abnormity detection method based on quality control chart theory | |
| CN111855219A (en) | Diesel engine lubricating oil entering security parameter prediction method based on grey theory | |
| CN114936520A (en) | Steam turbine fault early warning analysis method based on improved MSET | |
| CN111563685B (en) | Power generation equipment state early warning method based on auto-associative kernel regression algorithm | |
| CN112576454A (en) | Wind turbine generator main shaft temperature early warning method and device based on multi-dimensional early warning strategy | |
| CN111192163B (en) | Generator reliability medium-short term prediction method based on wind turbine generator operating data | |
| WO2019003703A1 (en) | Diagnostic device and diagnostic method | |
| CN110826600B (en) | Engine surge prediction method based on adaptive resonance network online incremental learning | |
| CN112067289A (en) | Motor shaft and transmission shaft abnormal vibration early warning algorithm based on neural network | |
| CN115422714A (en) | Knowledge condition hybrid driving running state monitoring method for gas turbine | |
| CN116146515A (en) | Abnormality detection method for centrifugal fan equipment of cement plant | |
| Souza et al. | Evaluation of data based normal behavior models for fault detection in wind turbines | |
| US20180087489A1 (en) | Method for windmill farm monitoring | |
| CN114295367A (en) | Wind turbine generator gearbox working condition online monitoring method | |
| Tutiv'en et al. | Wind turbine main bearing condition monitoring via convolutional autoencoder neural networks |
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 |