CN115539220A - Fault detection method for temperature sensor of gas turbine blade channel - Google Patents
Fault detection method for temperature sensor of gas turbine blade channel Download PDFInfo
- Publication number
- CN115539220A CN115539220A CN202211220744.5A CN202211220744A CN115539220A CN 115539220 A CN115539220 A CN 115539220A CN 202211220744 A CN202211220744 A CN 202211220744A CN 115539220 A CN115539220 A CN 115539220A
- Authority
- CN
- China
- Prior art keywords
- data
- gas turbine
- blade channel
- temperature
- fault
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 15
- 238000000034 method Methods 0.000 claims abstract description 32
- 230000008859 change Effects 0.000 claims abstract description 12
- 238000004458 analytical method Methods 0.000 claims abstract description 10
- 230000009897 systematic effect Effects 0.000 claims abstract description 5
- 238000012216 screening Methods 0.000 claims abstract description 4
- 238000002485 combustion reaction Methods 0.000 claims description 15
- 238000004422 calculation algorithm Methods 0.000 claims description 9
- 230000002159 abnormal effect Effects 0.000 claims description 8
- 238000012549 training Methods 0.000 claims description 8
- 238000013528 artificial neural network Methods 0.000 claims description 7
- 238000012360 testing method Methods 0.000 claims description 6
- 239000000446 fuel Substances 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 4
- 230000009467 reduction Effects 0.000 claims description 4
- 238000005070 sampling Methods 0.000 claims description 4
- 238000000605 extraction Methods 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 238000010926 purge Methods 0.000 claims 1
- 238000003745 diagnosis Methods 0.000 abstract description 2
- 239000007789 gas Substances 0.000 description 48
- 238000005259 measurement Methods 0.000 description 7
- 239000002245 particle Substances 0.000 description 6
- 230000008901 benefit Effects 0.000 description 5
- 238000001914 filtration Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 238000009826 distribution Methods 0.000 description 3
- 238000004140 cleaning Methods 0.000 description 2
- 238000003066 decision tree Methods 0.000 description 2
- 238000011478 gradient descent method Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- VNWKTOKETHGBQD-UHFFFAOYSA-N methane Chemical compound C VNWKTOKETHGBQD-UHFFFAOYSA-N 0.000 description 2
- 238000010248 power generation Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- OKTJSMMVPCPJKN-UHFFFAOYSA-N Carbon Chemical compound [C] OKTJSMMVPCPJKN-UHFFFAOYSA-N 0.000 description 1
- 230000002547 anomalous effect Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000009529 body temperature measurement Methods 0.000 description 1
- 229910052799 carbon Inorganic materials 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 239000003345 natural gas Substances 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000013024 troubleshooting Methods 0.000 description 1
Images
Classifications
-
- F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
- F02—COMBUSTION ENGINES; HOT-GAS OR COMBUSTION-PRODUCT ENGINE PLANTS
- F02C—GAS-TURBINE PLANTS; AIR INTAKES FOR JET-PROPULSION PLANTS; CONTROLLING FUEL SUPPLY IN AIR-BREATHING JET-PROPULSION PLANTS
- F02C9/00—Controlling gas-turbine plants; Controlling fuel supply in air- breathing jet-propulsion plants
-
- F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
- F01—MACHINES OR ENGINES IN GENERAL; ENGINE PLANTS IN GENERAL; STEAM ENGINES
- F01K—STEAM ENGINE PLANTS; STEAM ACCUMULATORS; ENGINE PLANTS NOT OTHERWISE PROVIDED FOR; ENGINES USING SPECIAL WORKING FLUIDS OR CYCLES
- F01K23/00—Plants characterised by more than one engine delivering power external to the plant, the engines being driven by different fluids
- F01K23/02—Plants characterised by more than one engine delivering power external to the plant, the engines being driven by different fluids the engine cycles being thermally coupled
- F01K23/06—Plants characterised by more than one engine delivering power external to the plant, the engines being driven by different fluids the engine cycles being thermally coupled combustion heat from one cycle heating the fluid in another cycle
- F01K23/10—Plants characterised by more than one engine delivering power external to the plant, the engines being driven by different fluids the engine cycles being thermally coupled combustion heat from one cycle heating the fluid in another cycle with exhaust fluid of one cycle heating the fluid in another cycle
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/088—Non-supervised learning, e.g. competitive learning
-
- F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
- F05—INDEXING SCHEMES RELATING TO ENGINES OR PUMPS IN VARIOUS SUBCLASSES OF CLASSES F01-F04
- F05D—INDEXING SCHEME FOR ASPECTS RELATING TO NON-POSITIVE-DISPLACEMENT MACHINES OR ENGINES, GAS-TURBINES OR JET-PROPULSION PLANTS
- F05D2260/00—Function
- F05D2260/80—Diagnostics
-
- F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
- F05—INDEXING SCHEMES RELATING TO ENGINES OR PUMPS IN VARIOUS SUBCLASSES OF CLASSES F01-F04
- F05D—INDEXING SCHEME FOR ASPECTS RELATING TO NON-POSITIVE-DISPLACEMENT MACHINES OR ENGINES, GAS-TURBINES OR JET-PROPULSION PLANTS
- F05D2260/00—Function
- F05D2260/81—Modelling or simulation
-
- F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
- F05—INDEXING SCHEMES RELATING TO ENGINES OR PUMPS IN VARIOUS SUBCLASSES OF CLASSES F01-F04
- F05D—INDEXING SCHEME FOR ASPECTS RELATING TO NON-POSITIVE-DISPLACEMENT MACHINES OR ENGINES, GAS-TURBINES OR JET-PROPULSION PLANTS
- F05D2270/00—Control
- F05D2270/30—Control parameters, e.g. input parameters
- F05D2270/303—Temperature
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Combustion & Propulsion (AREA)
- Chemical & Material Sciences (AREA)
- Theoretical Computer Science (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Mechanical Engineering (AREA)
- Control Of Turbines (AREA)
Abstract
The invention belongs to the field of state fault diagnosis and detection of a gas-steam combined cycle generator set, and particularly relates to a fault detection method for a gas turbine blade channel temperature sensor, which comprises the steps of carrying out systematic and comprehensive control logic analysis on the gas turbine blade channel temperature to obtain main thermal parameters influencing the temperature change of the gas turbine blade channel; collecting and screening fault-free data containing the whole operating conditions of the unit from a gas-steam combined cycle unit database; on the basis of the established LightGBM model and the self-encoder model, on one hand, the influence of normal changes of atmospheric temperature and working conditions on the temperature of a gas turbine blade channel is successfully eliminated, on the other hand, on the basis of the established sensor fault criterion, thermal parameter deviation caused by sensor faults and gas turbine component faults is successfully distinguished, real-time online fault detection on the gas turbine blade channel temperature sensor is realized, and the method has important significance for determining the operation state of a gas-steam combined cycle unit.
Description
Technical Field
The invention belongs to the field of state fault diagnosis and detection of a gas-steam combined cycle generator set, and particularly relates to a fault detection method for a temperature sensor of a gas turbine blade channel.
Background
The gas-steam combined cycle unit is an important power generation mode for constructing a novel power system, and has the advantages of flexible operation, quick start and stop, low carbon emission intensity, wide output power range and the like. In recent years, the loading amount of natural gas power generation continues to increase at a high speed, and the loading scale reaches more than 1.5 hundred million kilowatts in 2025 years. The gas turbine is used as a core device of the gas-steam combined cycle unit, the working environment of the gas turbine is complex, the working conditions are variable, and the failure risk of the gas turbine is greatly increased along with the increase of the running time. The temperature of the gas turbine blade channel is an important thermal parameter of the gas turbine, and is indispensable for accurately judging the state of the gas turbine. Therefore, the sensor is required to accurately obtain the true temperature of the blade passage. Once the sensor fails, an erroneous monitoring signal is generated, so that centralized control operators make a misjudgment on the running condition of the fuel machine, and even non-stop accidents may be caused.
When the gas turbine operates in a normal state, the temperature of the blade channel of the gas turbine changes along with the change of the atmospheric temperature and the working condition, and the change range is large, so whether the state of the sensor is normal or not is detected, firstly, the influence of the temperature of the blade channel along with the normal change of the atmospheric temperature and the working condition is processed and eliminated, and secondly, the influence of the fault of the gas turbine body equipment on the temperature of the blade channel is eliminated. At present, the temperature variation condition of a gas turbine blade channel is analyzed mainly by a method based on the combination of a physical model and simulation, but at the present stage, the heavy-duty gas turbine industry in China is mainly monopolized by foreign large complete machine manufacturing enterprises, the core key technology is lacked, the establishment of an accurate mathematical model is very difficult, and the calculation accuracy of the variation condition cannot be ensured.
Disclosure of Invention
The invention aims to overcome the defects in the prior art and provides a fault detection method for a temperature sensor of a blade channel of a gas turbine.
In order to achieve the purpose, the invention adopts the following technical scheme:
a method for detecting faults of a temperature sensor of a blade channel of a gas turbine comprises the following steps:
step 1, carrying out systematic and comprehensive control logic analysis on the temperature of a gas turbine blade channel to obtain main thermal parameters influencing the temperature change of the gas turbine blade channel;
step 2, collecting and screening fault-free data containing the whole operating conditions of the unit from a gas-steam combined cycle unit database;
step 3, in order to ensure the accuracy of the data, the situations that the measured data collected in the step 2 may have null values, abnormal distortion and the like are considered, so that the collected data are cleaned and filtered;
step 4, establishing a regression model of blade channel temperature control theoretical values according to historical fault-free full-working-condition operation data based on a Light Gradient Boosting Machine (Light GBM) method, and training and testing the model by using the data in the step 3;
step 5, establishing a blade channel temperature coding-decoding state reduction regression model according to historical fault-free full-working-condition operation data based on a self-encoder method, and training and testing the model by using the data in the step 3;
step 6, mutual analysis is carried out on the temperature residual errors of the blade channel based on the LightGBM model in the step 4 and the self-encoder model in the step 5, and corresponding threshold values are determined;
step 7, establishing a fault judgment criterion of the gas turbine blade channel temperature sensor;
and 8, utilizing the actually measured operation data of the gas turbine and based on the fault judgment criterion in the step 7, realizing real-time online fault detection on the gas turbine blade channel temperature sensor.
Preferably, the main thermal parameters influencing the temperature change of the gas turbine blade passage in the step 1 are mainly as follows: the shell pressure ratio of the combustion chamber, the inlet air temperature of the air compressor, the fuel flow, the opening degree of a bypass valve of the combustion chamber, the IGV opening degree, the pressure of the combustion chamber, the power of the gas turbine and the like; the calculation formula of the shell pressure ratio of the combustion chamber is as follows:
in the formula: PCS is compressor outlet pressure, and P0 is ambient atmospheric pressure.
Preferably, in the step 2, fault-free data containing the full operating conditions of the unit are collected and screened from a gas-steam combined cycle unit database; the acquisition time period is one year, and the data sampling frequency is 5 s-1 min; collecting data points includes: vane channel temperatures #1 to #20.
Preferably, in order to ensure the accuracy of the data in step 3, the collected data is cleaned and filtered in consideration of the fact that the measurement data collected in step 2 may have null values, abnormal distortion, and the like.
Preferably, the cleaned data comprises null data and outlier data, and further comprises data of a shutdown working condition, namely data corresponding to a time period when the unit load is 0 or close to 0;
null data is data with null values at one or more measuring points at a certain moment, and outlier data is data beyond a normal range; writing corresponding codes to remove null data.
Preferably, the Light Gradient Boosting Machine model in the step 4 is a framework for implementing a GBDT (Gradient Boosting Decision Tree) algorithm; the LightGBM model input layer has 3 dimensions, and comprises the active power of a gas turbine, the shell pressure ratio of a combustion chamber and the inlet air temperature of a gas compressor; the output layer has 20 dimensions and comprises blade channel temperatures #1 to #20. The blade channel temperatures #1 to #20 output by the LightGBM model are called theoretical control blade channel temperaturesDegrees #1 to #20, expressed as TCT i i＝1,2,3……20。
Preferably, the Auto-Encoder (AE) in step 5 is an unsupervised learning model; based on a back propagation algorithm and an optimization method (such as a gradient descent method), the input data T is used as supervision to guide the neural network to try to learn a mapping relation, and therefore a reconstructed output T is obtained R (ii) a With the aid of the nonlinear feature extraction capability of deep neural networks, good data representation can be obtained from the encoder.
Preferably, in the step 6, the blade channel temperature residuals are mutually analyzed based on the LightGBM model in the step 4 and the self-encoder model in the step 5, and the corresponding threshold is determined.
Preferably, the step 7 establishes a fault judgment criterion of the gas turbine blade passage temperature sensor, wherein the criterion is as follows:
ε 1i ≥threshold LGB
ε 2i ≥threshold AE
ε 3i ≤threshold LGB-AE
compared with the prior art, have following advantage:
the invention provides a fault detection method for a gas turbine blade channel temperature sensor, which aims at solving the problems that thermal parameter factors influencing the temperature of a gas turbine blade channel are numerous and cannot be accurately determined, carries out systematic and comprehensive control logic analysis on the temperature of the gas turbine blade channel, and extracts main thermal parameters influencing the temperature change of the gas turbine blade channel from a mechanism angle. On the basis of cleaning and preprocessing a large amount of historical data, a LightGBM model is established for the temperature of a gas turbine blade channel, the advantages of low memory consumption, distributed support and the like are utilized, massive data can be rapidly processed, and the influence of normal change of atmospheric temperature and load working conditions on the temperature of the gas turbine blade channel is eliminated by utilizing the idea of 'mechanism + big data'. On the other hand, a self-encoder model is established for the temperature of the blade channel, and the state reconstruction of the temperature of the blade channel under the fault-free condition is realized from the idea of big data. And through residual analysis and the established sensor fault criterion, thermal parameter deviation caused by sensor faults and gas turbine component faults is successfully distinguished, real-time online gas turbine blade channel temperature sensor fault detection is achieved, specific thermocouple faults can be accurately positioned, and workers can obtain more detailed fault information and timely troubleshooting is facilitated.
Drawings
FIG. 1 is a schematic flow diagram of a method for fault detection of a gas turbine blade passage temperature sensor in accordance with the present invention;
FIG. 2 is a schematic diagram of a gas turbine operating process and sensor measurement point distribution;
FIG. 3 is a schematic diagram of a regression model of theoretical values of temperature control for a blade passage;
FIG. 4 is a schematic diagram of a blade channel temperature encoding-decoding state reduction regression model.
Detailed Description
In order that the above objects, features and advantages of the present invention will be readily understood and appreciated, a more particular description of the invention briefly described above will be rendered by reference to specific embodiments thereof which are illustrated in the appended drawings.
Step 1, carrying out systematic and comprehensive control logic analysis on the temperature of the gas turbine blade channel, and obtaining main thermal parameters influencing the temperature change of the gas turbine blade channel.
And 2, collecting and screening fault-free data containing the full operating conditions of the unit from a gas-steam combined cycle unit database.
And 3, in order to ensure the accuracy of the data, the collected data is cleaned and filtered in consideration of the fact that the measured data collected in the step 2 may have null values, abnormal distortion and the like.
And 4, establishing a blade channel temperature control theoretical value regression model according to historical fault-free full-working-condition operation data based on a Light Gradient Boosting Machine (LightGBM) method, as shown in the figure 3, and training and testing the model by using the data in the step 3.
And 5, establishing a blade channel temperature coding-decoding state reduction regression model according to historical fault-free full-working-condition operation data by using a self-encoder-based method, as shown in FIG. 4, and training and testing the model by using the data in the step 3.
And 6, mutually analyzing the temperature residual errors of the blade channel based on the LightGBM model in the step 4 and the self-encoder model in the step 5, and determining corresponding threshold values.
And 7, establishing a fault judgment criterion of the gas turbine blade channel temperature sensor.
And 8, realizing real-time online fault detection of the gas turbine blade channel temperature sensor by using the actually measured operation data of the gas turbine based on the fault judgment criterion in the step 7.
The main thermal parameters influencing the temperature change of the blade channel of the gas turbine in the step 1 mainly comprise: the combustor shell pressure ratio, the compressor inlet air temperature, the fuel flow, the combustor bypass valve opening, the IGV opening, the combustor pressure, the gas turbine power, etc., are specifically shown in table 1. The calculation formula of the shell pressure ratio of the combustion chamber is as follows:
in the formula: PCS is compressor outlet pressure, and P0 is ambient atmospheric pressure.
TABLE 1 Primary thermal parameters affecting gas turbine blade passage temperature variation
And in the step 2, non-fault data containing the full operating conditions of the unit are collected and screened from a gas-steam combined cycle unit database. The recommended acquisition time period is one year, and the data sampling frequency is 5 s-1 min. Collecting data points includes: blade channel temperatures #1 to #20 (generally, 20 thermocouple sensors are circumferentially distributed in a blade channel of a combustion engine for temperature measurement), ambient atmospheric pressure, compressor inlet air temperature, compressor outlet pressure, fuel flow, opening degree of a combustion chamber bypass valve, IGV opening degree, combustion chamber pressure and gas turbine power. The gas turbine work flow and the sensor measuring point distribution are schematically shown in FIG. 2.
In the step 3, in order to ensure the accuracy of the data, the situations that the measured data acquired in the step 2 may have null values, abnormal distortion and the like are considered, so that the acquired data is cleaned and filtered. The cleaning data can comprise null value data and outlier data, and further can comprise data of shutdown working conditions, namely data corresponding to a time period when the unit load is 0 or is close to 0.
The null data is data of null values at one or more measuring points at a certain moment, and the outlier data is data beyond a normal range. Writing corresponding codes to remove null data, and removing outlier data by adopting a box-line graph method. Suppose q 1 、q 3 For the 1 st quartile, 3 rd quartile of data, the box plot method may be expressed as:
x max ＝q 3 +1.5×(q 3 -q 1 )
x min ＝q 3 -1.5×(q 3 -q 1 )
wherein x is max Representing abnormal maxima, x, in the data min Is an anomalous minimum in the data. And if the data is smaller than the abnormal minimum value and larger than the abnormal maximum value, determining the data as outlier data, and removing the outlier data.
Then, the data is subjected to filtering processing.
The method of filtering the data may be a particle filter method. Particle Filters (PF) algorithm is one of the key methods to solve the problem of parameter estimation and state filtering of a nonlinear non-gaussian dynamic system, and has no limitation on the process noise and measurement noise of the system, and the accuracy of the algorithm can approach the optimal estimation. Recursively estimating a posterior probability density p (x) of a nonlinear system state from noisy observations 0：k |z 1：k ). Wherein x is 0：k ＝{x 0 ，x 1 ，…x k Denotes the sequence of states produced by the system at time k, z 1：k ＝{z 1 ，z 2 ，…z k Denotes the observation sequence.
The basic idea of particle filtering is to construct a sample-based posterior probability density function using a set of N particles
Posterior probability density p (x) representing system state 0：k |z 1：k )。
Wherein, { x 0：k I =0,1, …, N } represents a set of support samples (particles), extracted from the state space of the posterior probability distribution. Each sample particle has a weight of
And satisfy
And finally completing the filtering process of the data by continuously updating the weight value based on selecting a proper importance sampling density function.
The Light gbm (Light Gradient Boosting Machine) model in the step 4 is a framework for implementing the GBDT (Gradient Boosting Decision Tree) algorithm, supports high-efficiency parallel training, and has the advantages of higher training speed, lower memory consumption, better accuracy, support of distributed type, capability of rapidly processing mass data, and the like. The lightGBM model input layer has 3 dimensions, and comprises the active power of a gas turbine, the shell pressure ratio of a combustion chamber and the inlet air temperature of a gas compressor; the output layer has 20 dimensions and comprises blade channel temperatures #1 to #20. The blade channel temperatures # 1- #20 output by the LightGBM model are called theoretical controlled blade channel temperatures # 1- #20 and are denoted as TCT i i＝1,2,3……20。
The Auto-Encoder (AE) in step 5 is an unsupervised learning model. Based on back propagation algorithm and optimization method (such as gradient descent method)) The input data T itself is used as supervision to guide the neural network to try to learn a mapping relation, so as to obtain a reconstructed output T R . With the aid of the nonlinear feature extraction capability of deep neural networks, good data representation can be obtained from the encoder. The self-encoder algorithm model contains two main parts: an Encoder and a Decoder.
The encoder is used for encoding the high-dimensional input T into the low-dimensional hidden variable h, so that the neural network is forced to learn the characteristics with the most information quantity, and the encoding process is as follows:
h＝g θ1 (T)＝σ(W 1 T+b 1 )
the decoder is used for restoring the hidden variable h of the hidden layer to the original dimension, and the best state is that the output of the decoder can perfectly or approximately restore the original input, namely T R ≈T
T R ＝g θ2 (h)＝σ(W 2 h+b 2 )
The input layer of the self-encoder model has 20 dimensions, and the input is the temperature of the blade channel from #1 to #20; the output layer of the self-encoder model has 20 dimensions, and the blade channel temperatures are from #1 to #20. The vane channel temperatures #1 to #20 output from the encoder model are referred to as state-restored vane channel temperatures #1 to #20. Denoted as SRT i i＝1,2,3……20
In the step 6, mutual analysis is performed on the blade channel temperature residual error based on the LightGBM model in the step 4 and the self-encoder model in the step 5, and a corresponding threshold value is determined. The specific residual formula is:
ε 1i ＝|TCT i -T i |,i＝1,2,3……20
ε 2i ＝|SRT i -T i |,i＝1,2,3……20
ε 3i ＝|TCT i -SRT i |,i＝1,2,3……20
in the formula: epsilon 1i Residual error, epsilon, of thermocouple measurement data of blade channel temperature for LightGBM model output 2i The output of the self-encoder model has a residual error epsilon with the thermocouple measurement data of the blade channel temperature 3i Is LighResidual error of tGBM model output and self-encoder model output data, T i Data values were measured for the blade channel temperature thermocouples. Wherein the LightGBM model output has a residual threshold, referred to herein as threshold, from thermocouple measurement data of blade passageway temperature LGB (ii) a There is a residual threshold, referred to herein as threshold, from the encoder model output and the blade channel temperature thermocouple measurement data AE (ii) a The LightGBM model output has a residual threshold, referred to herein as threshold, from the encoder model output LGB-AE 。
By making pairs of epsilon 1i 、ε 2i 、ε 3i The time series is statistically analyzed to determine the corresponding threshold. The method adopts a nuclear density estimation analysis method, and the expression of the nuclear density is as follows:
in the formula:
for the estimated probability density value, n is the number of samples, h is the window width, and K is the kernel function. The selected confidence is 99.5%, and a corresponding threshold is obtained by adopting a Gaussian kernel function.
Step 7, establishing a fault judgment criterion of the gas turbine blade channel temperature sensor, wherein the criterion is as follows:
ε 1i ≥threshold LGB
ε 2i ≥threshold AE
ε 3i ≤threshold LGB-AE
while the foregoing is directed to embodiments of the present invention, it will be appreciated by those skilled in the art that various changes may be made in the embodiments without departing from the principles of the invention, and that such changes and modifications are intended to be included within the scope of the appended claims.
Claims (9)
1. A fault detection method for a temperature sensor of a blade channel of a gas turbine is characterized by comprising the following steps:
step 1, carrying out systematic and comprehensive control logic analysis on the temperature of a gas turbine blade channel to obtain main thermal parameters influencing the temperature change of the gas turbine blade channel;
step 2, collecting and screening fault-free data containing the whole operating conditions of the unit from a gas-steam combined cycle unit database;
step 3, in order to ensure the accuracy of the data, the situations that the measured data collected in the step 2 may have null values, abnormal distortion and the like are considered, so that the collected data are cleaned and filtered;
step 4, establishing a regression model of the blade channel temperature control theoretical value according to historical fault-free full-working-condition operation data based on a LightGBM model method, and training and testing the model by using the data in the step 3;
step 5, establishing a blade channel temperature coding-decoding state reduction regression model according to historical fault-free full-working-condition operation data based on a self-encoder method, and training and testing the model by using the data in the step 3;
step 6, mutual analysis is carried out on the temperature residual errors of the blade channel based on the LightGBM model in the step 4 and the self-encoder model in the step 5, and corresponding threshold values are determined;
step 7, establishing a fault judgment criterion of a gas turbine blade channel temperature sensor;
and 8, utilizing the actually measured operation data of the gas turbine and based on the fault judgment criterion in the step 7, realizing real-time online fault detection on the gas turbine blade channel temperature sensor.
2. The method of detecting a fault in a turbine blade path temperature sensor of claim 1, wherein the thermal parameters affecting the temperature change of the turbine blade path in step 1 are: the method comprises the following steps of (1) controlling the shell pressure ratio of a combustion chamber, the inlet air temperature of a gas compressor, the fuel flow, the opening degree of a bypass valve of the combustion chamber, the opening degree of an IGV (insulated Gate Bipolar translator), the pressure of the combustion chamber, the power of a gas turbine and the like; the calculation formula of the shell pressure ratio of the combustion chamber is as follows:
in the formula: PCS is compressor outlet pressure, and P0 is ambient atmospheric pressure.
3. The method for detecting the fault of the gas turbine blade channel temperature sensor according to claim 2, wherein in the step 2, fault-free data containing the full operating condition of the unit is collected and screened from a gas-steam combined cycle unit database; the acquisition time period is one year, and the data sampling frequency is 5 s-1 min; collecting data points includes: vane channel temperatures #1 to #20.
4. The method of claim 3, wherein the collected data is cleaned and filtered.
5. The method of claim 4, wherein the purge data includes null data and outlier data, and further includes data of shutdown condition, i.e. data corresponding to a time period when the unit load is 0 or close to 0;
null data is data with null values at one or more measuring points at a certain moment, and outlier data is data beyond a normal range; writing corresponding codes to remove null data.
6. The method of claim 1, wherein the LightGBM model in step 4 is a framework for implementing GBDT algorithm; the LightGBM model input layer has 3 dimensions, and comprises the active power of a gas turbine, the shell pressure ratio of a combustion chamber and the inlet air temperature of a gas compressor; the output layer is 20-dimensional and comprises the temperature of a blade channel from #1 to #20; the blade channel temperatures # 1- #20 output by the LightGBM model are called theoretical control blade channel temperatures # 1- #20, expressed as TCT i i＝1,2,3……20。
7. The method of claim 1, wherein said step 5 self-encoder is an unsupervised learning model; based on a back propagation algorithm and an optimization method, the input data T is used as supervision to guide the neural network to try to learn a mapping relation, and therefore a reconstructed output T is obtained R (ii) a With the aid of the nonlinear feature extraction capability of deep neural networks, good data representation can be obtained from the encoder.
8. The method of claim 1, wherein in step 6, the blade path temperature residuals are analyzed with each other based on the LightGBM model in step 4 and the self-encoder model in step 5 to determine the corresponding threshold.
9. The method for detecting the fault of the temperature sensor of the blade channel of the gas turbine as claimed in claim 1, wherein the step 7 establishes the fault judgment criteria of the temperature sensor of the blade channel of the gas turbine, and the criteria are as follows:
ε 1i ≥threshold LGB
ε 2i ≥threshold AE
ε 3i ≤threshold LGB-AE 。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211220744.5A CN115539220B (en) | 2022-10-08 | 2022-10-08 | Gas turbine blade channel temperature sensor fault detection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211220744.5A CN115539220B (en) | 2022-10-08 | 2022-10-08 | Gas turbine blade channel temperature sensor fault detection method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115539220A true CN115539220A (en) | 2022-12-30 |
| CN115539220B CN115539220B (en) | 2024-07-05 |
Family
ID=84731754
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211220744.5A Active CN115539220B (en) | 2022-10-08 | 2022-10-08 | Gas turbine blade channel temperature sensor fault detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115539220B (en) |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090228230A1 (en) * | 2008-03-06 | 2009-09-10 | General Electric Company | System and method for real-time detection of gas turbine or aircraft engine blade problems |
| CN101694182A (en) * | 2009-09-29 | 2010-04-14 | 上海中科清洁能源技术发展中心 | On-line failure diagnosis, prediction and feedback control method of small/medium size gas turbine and device thereof |
| KR20140037347A (en) * | 2012-09-17 | 2014-03-27 | 이봉우 | Engine analysis apparatus using oxygen sensor |
| CN110513336A (en) * | 2019-09-05 | 2019-11-29 | 哈尔滨电气股份有限公司 | A kind of determination method of power station gas turbine off-line washing time |
| US20190379589A1 (en) * | 2018-06-12 | 2019-12-12 | Ciena Corporation | Pattern detection in time-series data |
| US20200141653A1 (en) * | 2018-11-02 | 2020-05-07 | Honeywell International Inc. | Flame analytics system |
| CN111441869A (en) * | 2020-03-29 | 2020-07-24 | 至玥腾风科技集团有限公司 | Method and system for starting micro gas turbine |
| US20200387797A1 (en) * | 2018-06-12 | 2020-12-10 | Ciena Corporation | Unsupervised outlier detection in time-series data |
| CN113834656A (en) * | 2021-08-27 | 2021-12-24 | 西安电子科技大学 | Bearing fault diagnosis method, system, equipment and terminal |
| CN113919237A (en) * | 2021-11-02 | 2022-01-11 | 南京凯奥思数据技术有限公司 | Method for online working condition segmentation and fault diagnosis of fan equipment |
-
2022
- 2022-10-08 CN CN202211220744.5A patent/CN115539220B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090228230A1 (en) * | 2008-03-06 | 2009-09-10 | General Electric Company | System and method for real-time detection of gas turbine or aircraft engine blade problems |
| CN101694182A (en) * | 2009-09-29 | 2010-04-14 | 上海中科清洁能源技术发展中心 | On-line failure diagnosis, prediction and feedback control method of small/medium size gas turbine and device thereof |
| KR20140037347A (en) * | 2012-09-17 | 2014-03-27 | 이봉우 | Engine analysis apparatus using oxygen sensor |
| US20190379589A1 (en) * | 2018-06-12 | 2019-12-12 | Ciena Corporation | Pattern detection in time-series data |
| US20200387797A1 (en) * | 2018-06-12 | 2020-12-10 | Ciena Corporation | Unsupervised outlier detection in time-series data |
| US20200141653A1 (en) * | 2018-11-02 | 2020-05-07 | Honeywell International Inc. | Flame analytics system |
| CN110513336A (en) * | 2019-09-05 | 2019-11-29 | 哈尔滨电气股份有限公司 | A kind of determination method of power station gas turbine off-line washing time |
| CN111441869A (en) * | 2020-03-29 | 2020-07-24 | 至玥腾风科技集团有限公司 | Method and system for starting micro gas turbine |
| CN113834656A (en) * | 2021-08-27 | 2021-12-24 | 西安电子科技大学 | Bearing fault diagnosis method, system, equipment and terminal |
| CN113919237A (en) * | 2021-11-02 | 2022-01-11 | 南京凯奥思数据技术有限公司 | Method for online working condition segmentation and fault diagnosis of fan equipment |
Non-Patent Citations (2)
| Title |
|---|
| 于美玲;刘永文;: "基于非线性模型的燃气轮机热参数故障诊断及仿真", 热能动力工程, no. 06, 20 November 2011 (2011-11-20) * |
| 刘钰宸;安静;: "一种基于lightGBM框架改进的GBDT风力发电机叶片开裂预测方法", 应用技术学报, no. 01, 30 March 2020 (2020-03-30) * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115539220B (en) | 2024-07-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111159844B (en) | Abnormity detection method for exhaust temperature of gas turbine of power station | |
| CN111539553B (en) | Wind turbine generator fault early warning method based on SVR algorithm and off-peak degree | |
| CN112364567B (en) | Residual life prediction method based on consistency check of similarity of degraded tracks | |
| CN109443423B (en) | Gas turbine unit carbon emission data correlation degree inspection method based on product difference analysis | |
| CN105971738B (en) | A kind of method of gas turbine group flue gas flow on-line monitoring | |
| CN110779745A (en) | Heat exchanger early fault diagnosis method based on BP neural network | |
| CN114757380B (en) | Fault early warning system and method for thermal power plant, electronic equipment and storage medium | |
| CN112418306B (en) | Gas turbine compressor fault early warning method based on LSTM-SVM | |
| Togni et al. | A combined technique of Kalman filter, artificial neural network and fuzzy logic for gas turbines and signal fault isolation | |
| CN110513336B (en) | Method for determining offline water washing time of gas turbine of power station | |
| CN111275570A (en) | Wind turbine generator set power abnormal value detection method based on iterative statistics and hypothesis test | |
| CN105930629A (en) | On-line fault diagnosis method based on massive amounts of operating data | |
| CN112327701B (en) | Slow characteristic network monitoring method for nonlinear dynamic industrial process | |
| CN112069457B (en) | Non-stationary dynamic process anomaly monitoring method based on dynamic stationary subspace analysis | |
| CN114326486A (en) | Process monitoring method based on probability slow feature analysis and elastic weight consolidation | |
| Wang et al. | Research on anomaly detection and positioning of marine nuclear power steam turbine unit based on isolated forest | |
| CN117829342A (en) | Coal-fired power plant nitrogen oxide emission prediction method based on improved random forest algorithm | |
| CN115539220B (en) | Gas turbine blade channel temperature sensor fault detection method | |
| CN108895839B (en) | sintering machine air leakage fault online diagnosis method | |
| CN108759479B (en) | Sintering machine air leakage fault online diagnosis method based on blind source separation unilateral detection method | |
| CN105279553A (en) | Method for identifying fault degree of high-pressure heater water supply system | |
| CN113431800A (en) | Method for distinguishing running state of fan in real time | |
| CN112051839A (en) | Process monitoring and fault diagnosis method based on tree structure sparsity | |
| CN108009580B (en) | Method for measuring humidity of coal at coal moisture control outlet | |
| CN114545899B (en) | Multi-sensor fault signal reconstruction method for gas turbine system based on priori knowledge |
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 |