CN110441065B - Gas turbine on-line detection method and device based on LSTM - Google Patents
Gas turbine on-line detection method and device based on LSTM Download PDFInfo
- Publication number
- CN110441065B CN110441065B CN201910601564.3A CN201910601564A CN110441065B CN 110441065 B CN110441065 B CN 110441065B CN 201910601564 A CN201910601564 A CN 201910601564A CN 110441065 B CN110441065 B CN 110441065B
- Authority
- CN
- China
- Prior art keywords
- gas turbine
- data
- lstm
- value
- training
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M15/00—Testing of engines
- G01M15/14—Testing gas-turbine engines or jet-propulsion engines
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Chemical & Material Sciences (AREA)
- Combustion & Propulsion (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
The invention belongs to the field of safety control systems of power plants, and particularly relates to an LSTM-based gas turbine online detection method and device. The method is characterized in that: the method comprises the following steps: collecting data; normalization processing; extracting characteristics; training an LSTM anomaly detection model; and (3) abnormal online detection, namely inputting data to be predicted into a trained detection model, obtaining a predicted value of the model, calculating the difference between the predicted value and the measured value of the sensor to obtain an absolute value, and judging that an abnormality occurs if the absolute value exceeds a given threshold value. The method is suitable for processing and predicting important events with relatively long intervals and delays in the time sequence, and is suitable for analyzing and fitting the time sequence. The concept and the technology of deep learning are fully utilized to automatically select hidden features in data information detected by an extraction equipment sensor, and then online anomaly detection based on real-time measuring point data of the gas turbine is realized. The invention has large data acquisition quantity, small analysis error and high early warning result accuracy.
Description
Technical Field
The invention belongs to the field of safety control systems of power plants, and particularly relates to an LSTM-based gas turbine online detection method and device.
Background
The gas turbine is one of the most important power machines at present, and has wide application in the fields of aviation, electric power, ships and the like. The gas turbine has a complex structure and a severe working environment, is easy to cause various faults, and can seriously affect the safety and reliability of the operation if the faults cannot be found and maintained in time. With the increase of the application amount of the gas turbine, people pay more and more attention to the working condition of the gas turbine, once the gas turbine breaks down and stops working, the stability of a power system is influenced, huge economic loss is caused, even the stable development of national economy is influenced, and the schedule is provided for the research of the gas turbine faults.
In the prior art, methods for detecting an abnormality in a gas turbine can be roughly classified into two types. One is a method adopting a mechanism model, a physical model is established based on the thermodynamic property and the thermodynamic principle of gas, each KPI index of the gas turbine is calculated by the model, and the KPI index is compared with an actual measurement value. If the measured value has a large deviation from the theoretical value, the gas turbine is considered to be abnormal. The main problem of the mechanism model is that when an analysis model is established by applying a physics principle, a large number of premise assumptions and simplification conditions exist, and the mechanism model is not suitable for a complex system under a real condition.
With the fact that the device detection data obtained in the device operation process are more and more diverse and the device structure and the operation environment are more and more complex, the operation mechanism of the device is fully understood, and the difficulty in extracting the data features related to the device fault state is higher and higher, for this reason, another type of anomaly detection technology tries to establish a mathematical model by using a data analysis and machine learning method, automatically and intelligently searches the mapping relation between the data features and the anomaly modes, and the accuracy of the anomaly detection method is improved. Common machine learning methods include fuzzy logic, Support Vector Machines (SVMs), artificial neural networks, and the like. The machine learning method can fully mine the data of the information, realize data driving to the maximum extent and reduce human intervention. However, because the gas turbine has short time for thermal power application, complex and various fault types and few repeated cases, the fault mechanism is difficult to understand sufficiently, so that it is difficult to extract and define relevant features for anomaly detection accurately, and it is difficult to ensure the accuracy of anomaly detection.
Chinese patent 201410745943.7 discloses a method for predicting the fault trend of a steam turbine with a self-adaptive quantum neural network. The method improves a traditional three-layer BP neural network model, introduces a quantum neural network, analyzes trend contribution force of different historical data in an input layer, strengthens influence of latest data on the trend, increases direct connection weight from the input layer to an output layer, and adaptively adjusts an excitation function according to signal characteristics in the output layer so as to improve convergence speed and prediction accuracy; a method of adaptive learning efficiency is introduced to improve convergence speed. The method has good reliability and robustness, is a key technical research for solving the problem of the steam turbine fault trend prediction, and can be widely applied to the steam turbine fault trend prediction. The method has the defects that the prediction error is large, and the subsequent control is greatly influenced.
Disclosure of Invention
Aiming at the problems in the prior art, the invention provides an LSTM-based gas turbine online detection method with small error and high accuracy and a device using the method.
The invention is realized in this way, a gas turbine online detection method based on LSTM, which is characterized in that: the method comprises the following steps:
the method comprises the following steps: acquiring data, namely acquiring real-time measuring point data from monitoring software of the gas turbine;
step two: normalization processing, namely unifying the dimension of the data of each measuring point;
step three: extracting characteristics, namely calculating the correlation among the normalized data sets, carrying out linear change on the normalized data characteristics, mapping the data from a high-dimensional space to a low-dimensional space, reducing the data dimension, and extracting main characteristics;
step four: training an LSTM anomaly detection model by taking the data processed in the step three as a training data set;
step five: and (3) abnormal online detection, namely inputting data to be predicted into a trained detection model, obtaining a predicted value of the model, calculating the difference between the predicted value and the measured value of the sensor to obtain an absolute value, and judging that an abnormality occurs if the absolute value exceeds a given threshold value.
In the second step, the data is normalized by calculating the mean value mu and the standard deviation sigma in the original data by a Z-score normalization method according to the following formula:
the data after the processing by the method conform to the standard normal distribution with the mean value of 0 and the standard deviation of 1.
The third step comprises the following steps:
step 3-1: calculating a correlation coefficient matrix, and inputting data { X ] of N measuring points1,X2,...,XNCalculating a Pearson correlation coefficient between every two measuring points according to a formula (2) to form a correlation coefficient matrix,
where rij (i, j ═ 1, 2.., n) denotes the original vector Xi，XjCorrelation coefficient of degree of correlation;
wherein r isijIs a one-dimensional vector XiAnd XjCorrelation coefficient of (2), XikRepresenting a one-dimensional vector XiThe k-th element of (1), XjkRepresenting a one-dimensional vector XjThe k-th element of (a) is,
representing a one-dimensional vector XiIs determined by the average value of (a) of (b),
representing a one-dimensional vector XjThe calculation formula is as follows:
step 3-2: calculating eigenvalue and eigenvector, solving the equation of | λ E-R | ═ 0 (where E is unit vector and R represents correlation coefficient matrix) to obtain eigenvalue, sorting according to size, and then respectively obtaining corresponding eigenvalue λiA feature vector U of (i ═ 1, 2.., n)i( i 1, 2.., n), the principal component matrix Y is calculated according to the following formula,
wherein, UijIs a characteristic value λiThe feature vector U corresponding to (i ═ 1, 2.. times.n)i＝[Ui1，Ui2，…Uin]The jth element in (a).
Step 3-3: calculating the information contribution rate of each eigenvector, and calculating the eigenvalue lambdaiA cumulative variance contribution rate cpv (cumulative percent variance) of (i ═ 1,2,. and n), which is expressed as follows:
the fourth step comprises the following steps:
step 4-1: establishing a neural network model, giving a data set after characteristic extraction:
wherein
Is a set of input sensor data vectors, yiTo predict the sensor value labels at the time of training, the LSTM neural network model is trained such that:
where Func denotes what the network model learns
A mapping to y;
step 4-2: selecting an error function, selecting a mean square error function, which is defined as follows:
wherein: m is the number of data in the data set after feature extraction, see step 4-1, yiIs a predictive sensor value tag entered in the data set,
is the sensor value predicted using the LSTM neural network model, see equation (6).
Step 4-3: determining an activation function, and adopting a tahn activation function to control the output within the range of [0,1] so as to protect and control the cell state; using softsign activation function at LSTM neuron input time;
step 4-4: training a neural network model, uniformly initializing all weight parameters in a range of [ -0.08,0.08] in an initial weight stage of the neural network so that the neural network model can remember all memories in the initial training stage, setting an initial offset value of an LSTM forgetting gate to be 1.0, setting initial values of an input gate and an output gate to be random floating point values in a [0,1] interval, then training the network by using a random gradient descent SGD, wherein the learning rate is 0.001, the attenuation factor is 0.95, training the model for 50 rounds, and multiplying the learning rate by the attenuation factor of 0.95 in each round after 10 rounds.
The LSTM-based gas turbine on-line detection device provided with the method is characterized by comprising a memory, a processor, I/O equipment and an alarm device which are electrically connected and store the program for realizing the method, wherein the I/O equipment is connected with a computer and/or a network provided with monitoring software of the gas turbine, and accesses and obtains real-time measuring point data.
The processor is connected with the handheld user end through wireless transmission.
The invention has the advantages and positive effects that:
the invention constructs an anomaly detection model based on the LSTM technology, is suitable for processing and predicting important events with relatively long intervals and delays in a time sequence, and is suitable for analyzing and fitting the time sequence. The concept and the technology of deep learning are fully utilized to automatically select hidden features in data information detected by an extraction equipment sensor, and then online anomaly detection based on real-time measuring point data of the gas turbine is realized. The invention has large data acquisition quantity, small analysis error and high early warning result accuracy.
Drawings
FIG. 1 is a flow block diagram of the present invention;
FIG. 2 is a schematic diagram of a neural network of the present invention;
FIG. 3 is a schematic diagram of the tanh and softsign functions of an embodiment of the present invention;
FIG. 4 shows the effect of detecting the abnormal temperature point of the compressor discharge air of the gas turbine according to the embodiment of the invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail with reference to the following embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
The following detailed description of the principles of the invention is provided in connection with the accompanying drawings.
Example 1:
as shown in fig. 1-3, the proposed LSTM-based gas turbine anomaly online detection method and apparatus includes the following steps:
the method comprises the following steps: and (6) data acquisition. The gas turbine mainly comprises three parts, namely a gas compressor, a combustion chamber and a turbine. The main survey point data is selected from the gas turbine monitoring software and analyzed as input to the LSTM neural network. The measuring point data comprises: GT IGV position (angle), GT IGV position, air humidity, gas turbine compressor inlet temperature, gas turbine inlet filter differential pressure measurement point (1) and measurement point (2), gas turbine inlet filter internal pressure, gas turbine inlet static pressure, GT IGV position-1 measurement point (1) and measurement point (2), gas turbine compressor inlet temperature measurement point (1), measurement point (2) and measurement point (3), gas turbine compressor outlet air temperature, gas turbine combustion chamber shell pressure measurement point (1), measurement point (2) and measurement point (3), gas turbine fuel A main pipe pressure, gas turbine fuel pilot main pipe pressure, main fuel main pipe pressure, gas turbine fuel top main pipe pressure, gas turbine combustion chamber bypass valve position, gas turbine blade channel highest temperature, gas turbine blade channel lowest temperature, gas turbine blade channel maximum temperature change, The method comprises the following steps of minimum temperature change of a gas turbine blade channel, maximum value of gas turbine exhaust temperature and minimum value of gas turbine exhaust temperature.
The data collected by the sensors are provided with a plurality of measuring points which have no essential difference, and the sensors are used for measuring the same index data, so that the correct reading and redundant measurement of a certain sensor after error is ensured.
Step 2: and (6) normalizing the data. For unifying the dimension of the data, the Z-score normalization method is used to calculate the mean value μ and the standard deviation σ in the raw data for normalization of the data, and the formula is as follows:
the data after the processing by the method conform to the standard normal distribution with the mean value of 0 and the standard deviation of 1.
And step 3: and (5) feature extraction. And calculating the correlation of the original data, and simultaneously carrying out linear change on the characteristics of the original data based on a principal component analysis method to extract main characteristics from the original data. In studying multivariate oriented statistical analysis problems, the more variables, the more computation and complexity that adds to the analysis problem. It is therefore desirable to discover and extract key variables in the course of performing quantitative analyses. The variables involved are small, but the amount of information contained is large enough. The principal component analysis utilizes the idea of dimension reduction, generates a series of comprehensive indexes which are not linearly related to each other by constructing proper linear combination of the original indexes, selects a few new comprehensive indexes from the comprehensive indexes, and enables the comprehensive indexes to contain information contained in the original indexes as much as possible, namely, the information of the original data is explained by using fewer indexes. The specific implementation method is that a given group of related variables are converted into another group of unrelated variables through a series of mathematical transformation, and the new variables are arranged according to the sequence that the variance is sequentially reduced. The method is characterized in that the total power of variables is kept unchanged in mathematical transformation, a first variable has the largest variance and is called a first principal component, and a second variable has the second largest variance and is not related to the first variable and is called a second principal component.
The characteristic extraction method based on principal component analysis comprises the following steps:
step 3-1: and calculating a correlation coefficient matrix. Inputting data of N measuring points { X1,X2,...,XNAnd calculating a Pearson correlation coefficient between every two measuring points according to a formula (2) to form a correlation coefficient matrix. Where rij (i, j ═ 1, 2.., n) denotes the original vector Xi，XjCorrelation coefficient of degree of correlation.
Wherein r isijIs a one-dimensional vector XiAnd XjCorrelation coefficient of (2), XikRepresenting a one-dimensional vector XiThe k-th element of (1), XjkRepresenting a one-dimensional vector XjThe k-th element of (a) is,
representing a one-dimensional vector XiIs determined by the average value of (a) of (b),
representing a one-dimensional vector XjThe calculation formula is as follows:
step 3-2: the eigenvalues and eigenvectors are computed. Firstly, solving a characteristic equation of lambda E-R0, wherein E is a unit vector, R represents a correlation coefficient matrix, solving characteristic values, sequencing according to the magnitude, and then respectively solving corresponding characteristic values lambdaiA feature vector U of (i ═ 1, 2.., n)i(i ═ 1, 2.., n). The principal component matrix Y is calculated according to the following formula.
Wherein, UijIs a characteristic value λiThe feature vector U corresponding to (i ═ 1, 2.. times.n)i＝[Ui1，Ui2，…Uin]The jth element in (a).
Step 3-3: calculating the information contribution rate of each eigenvector, and calculating the eigenvalue lambdaiA cumulative variance contribution rate cpv (cumulative percent variance) of (i ═ 1,2,. and n), which is expressed as follows:
and 4, step 4: an LSTM based fault detection model is trained. And training an LSTM neural network-based anomaly detection model based on the features extracted in the previous steps.
Step 4-1: and establishing a neural network model. And (4) performing model training by using supervised learning based on the feature extraction result of the step (3). Given the feature extracted data set:
wherein
Is a set of input sensor data vectors, yiIs a predictive sensor value label during training. The LSTM neural network model as shown in fig. 2 is trained such that:
where Func denotes what the network model learns
Mapping to y.
Step 4-2: an error function is selected. The neural network model needs to learn sensor data and then perform numerical prediction to assist in judging the operating state of the device, so that a Mean Squared Error (Mean Squared Error) Error function is selected, which is defined as follows:
wherein: m is the number of data in the data set after feature extraction, see step 4-1, yiIs a predictive sensor value tag entered in the data set,
is the sensor value predicted using the LSTM neural network model, see equation (6).
Step 4-3: an activation function is determined. The activation function of the gate control unit in the LSTM neural network adopts a tahn activation function to control the output within the range of [0,1] so as to protect and control the cell state; softsign activation functions are also used at the time of LSTM neuron input. the tahn activation function and softsign activation function are shown in fig. 3, and their definitions are shown in the following equations (8) and (9), respectively.
Step 4-4: and training the neural network model. In the initial weight stage of the neural network, all weight parameters are initialized uniformly in the range of [ -0.08,0.08], so that the model can remember all memories in the initial training stage, the initial bias value of the LSTM forgetting gate is set to be 1.0, and the initial values of the input gate and the output gate are random floating point values on the interval of [0,1 ]. The micro-batch stochastic gradient descent training network was then used with a learning rate of 0.001 and an attenuation factor of 0.95. Model 50 runs were trained and after 10 runs each learning rate was multiplied by a decay factor of 0.95.
And 5: and (4) abnormal online prediction. Given a set of input data
Outputting a predicted value according to the model M
And calculating a difference value d between the predicted value and the true value:
and setting a threshold value epsilon for judging the state of the equipment, if d is less than or equal to epsilon, considering that the equipment is in a normal running state, and if not, considering that the equipment is abnormal, and performing early warning.
FIG. 4 shows the anomaly detection effect of the gas turbine compressor discharge temperature measurement point. As indicated by the red circle in fig. 4, the lower line represents the predicted value of the model, and the upper line represents the measured value. As can be seen from the figure, a relatively large deviation exists between the model predicted value and the model measured value, and the equipment is considered to be abnormal at the position, so that early warning is carried out.
Example 2:
the LSTM-based gas turbine online detection device provided with the method comprises an electrically connected memory, a processor, I/O equipment and an alarm device, wherein the memory is used for storing a program for realizing the method, and the I/O equipment is connected with a computer and/or a network provided with monitoring software of the gas turbine and is used for accessing and acquiring real-time measuring point data.
The processor is connected with the handheld user end through wireless transmission. And remote monitoring and early warning are carried out through the handheld device.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents and improvements made within the spirit and principle of the present invention are intended to be included within the scope of the present invention.
Claims (6)
1. An LSTM-based gas turbine online detection method is characterized by comprising the following steps: the method comprises the following steps:
the method comprises the following steps: acquiring data, namely acquiring real-time measuring point data from monitoring software of the gas turbine; the measuring point data comprises: GT IGV position, air humidity, gas turbine compressor inlet air temperature, gas turbine inlet filter differential pressure measurement point, gas turbine inlet filter internal pressure, gas turbine inlet static pressure, gas turbine compressor inlet air temperature measurement point, gas turbine compressor outlet air temperature, gas turbine combustor casing pressure measurement point, gas turbine fuel A main pipe pressure, gas turbine fuel pilot main pipe pressure, gas turbine main fuel main pipe pressure, gas turbine fuel top main pipe pressure, gas turbine combustor bypass valve position, gas turbine blade channel maximum temperature, gas turbine blade channel minimum temperature, gas turbine blade channel maximum temperature change, gas turbine blade channel minimum temperature change, gas turbine exhaust temperature maximum, gas turbine exhaust temperature minimum;
step two: normalization processing, namely unifying the dimension of the data of each measuring point;
step three: extracting characteristics, namely calculating the correlation among the normalized data sets, carrying out linear change on the normalized data characteristics, mapping the data from a high-dimensional space to a low-dimensional space, reducing the data dimension, and extracting main characteristics;
step four: training an LSTM anomaly detection model by taking the data processed in the step three as a training data set; the method comprises the following steps:
step 4-1: establishing a neural network model, giving a data set after characteristic extraction:
wherein
Is a set of input sensor data vectors, yiTo predict the sensor value labels at the time of training, the LSTM neural network model is trained such that:
where Func denotes what the network model learns
A mapping to y;
step 4-2: selecting an error function, selecting a mean square error function, which is defined as follows:
wherein: m is the number of data in the data set after feature extraction, yiIs a predictive sensor value tag entered in the data set,
the sensor value is predicted by using an LSTM neural network model;
step 4-3: determining an activation function, and adopting a tahn activation function to control the output within the range of [0,1] so as to protect and control the cell state; using softsign activation function at LSTM neuron input time;
step 4-4: training a neural network model, uniformly initializing all weight parameters in the range of [ -0.08,0.08] in the weight initialization stage of the neural network so that the neural network model can remember all memories in the initial training stage, setting the initial offset value of an LSTM forgetting gate to be 1.0, setting the initial values of an input gate and an output gate to be random floating point values in the interval of [0,1], then using a random gradient to descend an SGD training network, wherein the learning rate is 0.001, the attenuation factor is 0.95, training the model for 50 rounds, and multiplying the learning rate by the attenuation factor of 0.95 in each round after 10 rounds;
step five: and (3) abnormal online detection, namely inputting data to be predicted into a trained detection model, obtaining a predicted value of the model, calculating the difference between the predicted value and the measured value of the sensor to obtain an absolute value, and judging that an abnormality occurs if the absolute value exceeds a given threshold value.
2. The LSTM-based gas turbine on-line testing method of claim 1, wherein the normalization of the data is performed by using Z-score normalization to calculate the mean μ and standard deviation σ of the raw data according to the following formula:
the data after the processing by the method conform to the standard normal distribution with the mean value of 0 and the standard deviation of 1.
3. The LSTM-based gas turbine online inspection method of claim 1, wherein said step three comprises the steps of:
step 3-1: calculating a correlation coefficient matrix, and inputting data { X ] of N measuring points1,X2,...,XNCalculating a Pearson correlation coefficient between every two measuring points according to a formula (2) to form a correlation coefficient matrix,
where rij (i, j ═ 1, 2.., n) denotes the original vector Xi，XjCorrelation coefficient of degree of correlation;
wherein r isijIs a one-dimensional vector XiAnd XjCorrelation coefficient of (2), XikRepresenting a one-dimensional vector XiThe k-th element of (1), XjkTo representOne-dimensional vector XjThe k-th element of (a) is,
representing a one-dimensional vector XiIs determined by the average value of (a) of (b),
representing a one-dimensional vector XjThe calculation formula is as follows:
step 3-2: calculating eigenvalue and eigenvector, solving the equation of | λ E-R | ═ 0, finding eigenvalue, sorting according to size, where E is unit vector, R represents correlation coefficient matrix, and finding corresponding eigenvalue λiA feature vector U of (i ═ 1, 2.., n)i(i 1, 2.., n), the principal component matrix Y is calculated according to the following formula,
wherein, UijIs a characteristic value λiThe feature vector U corresponding to (i ═ 1, 2.. times.n)i＝[Ui1,Ui2,…Uin]The jth element in (a);
step 3-3: calculating the information contribution rate of each eigenvector, and calculating the eigenvalue lambdaiA cumulative variance contribution rate cpv (cumulative percent variance) of (i ═ 1,2,. and n), which is expressed as follows:
4. the LSTM-based gas turbine online detection method of claim 1, wherein the gas turbine inlet filter differential pressure measurement point, the GT IGV position measurement point, the gas turbine compressor inlet temperature measurement point and the gas turbine combustor casing pressure measurement point are each provided with more than 2 measurement points.
5. An LSTM-based gas turbine on-line monitoring device for implementing the method of any one of claims 1-4, comprising an electrically connected memory storing a program for implementing the method, a processor, I/O equipment and an alarm device, wherein the I/O equipment is connected with a computer and/or a network for installing monitoring software of the gas turbine, accesses and obtains real-time station data.
6. The LSTM-based gas turbine online inspection device of claim 5, wherein the processor is connected to the handheld user end via wireless transmission.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910601564.3A CN110441065B (en) | 2019-07-04 | 2019-07-04 | Gas turbine on-line detection method and device based on LSTM |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910601564.3A CN110441065B (en) | 2019-07-04 | 2019-07-04 | Gas turbine on-line detection method and device based on LSTM |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110441065A CN110441065A (en) | 2019-11-12 |
| CN110441065B true CN110441065B (en) | 2022-02-08 |
Family
ID=68428472
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910601564.3A Active CN110441065B (en) | 2019-07-04 | 2019-07-04 | Gas turbine on-line detection method and device based on LSTM |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110441065B (en) |
Families Citing this family (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110874744B (en) * | 2019-11-18 | 2022-08-02 | 中国银联股份有限公司 | Data anomaly detection method and device |
| CN110926532A (en) * | 2019-11-29 | 2020-03-27 | 四川省生态环境科学研究院 | Digital monitoring system of city raise dust based on big data |
| CN110987448B (en) * | 2019-12-05 | 2021-12-21 | 潍柴动力股份有限公司 | Engine air inlet state monitoring method, device and equipment |
| CN111275288B (en) * | 2019-12-31 | 2023-12-26 | 华电国际电力股份有限公司十里泉发电厂 | XGBoost-based multidimensional data anomaly detection method and device |
| CN111272436B (en) * | 2020-02-11 | 2022-07-12 | 上海交通大学 | Gas turbine NOx prediction emission monitoring system and method |
| CN111985610B (en) * | 2020-07-15 | 2024-05-07 | 中国石油大学(北京) | Oil pumping well pump efficiency prediction system and method based on time sequence data |
| CN111931420B (en) * | 2020-08-07 | 2023-09-29 | 合肥工业大学 | Gas turbine fault prediction method based on nuclear regeneration Hilbert space |
| CN112115024B (en) * | 2020-09-03 | 2023-07-18 | 上海上讯信息技术股份有限公司 | Training method and device for fault prediction neural network model |
| CN112000015B (en) * | 2020-09-07 | 2022-08-05 | 华北电力大学 | Intelligent BIT design method for heavy-duty gas turbine control system controller module based on LSTM and bio-excitation neural network |
| CN112093025B (en) * | 2020-09-23 | 2022-12-09 | 江苏科技大学 | Control system and control method for variable air volume air distributor air valve |
| CN112668200B (en) * | 2021-01-06 | 2023-08-29 | 西安理工大学 | Special equipment safety analysis method and system |
| CN113033663A (en) * | 2021-03-26 | 2021-06-25 | 同济大学 | Automatic container terminal equipment health prediction method based on machine learning |
| CN113657628A (en) * | 2021-08-20 | 2021-11-16 | 武汉霖汐科技有限公司 | Industrial equipment monitoring method and system, electronic equipment and storage medium |
| CN114184389B (en) * | 2021-11-18 | 2023-08-15 | 中国船舶重工集团公司第七O三研究所无锡分部 | Cooling effectiveness inspection method for exhaust injection ventilation in marine gas turbine box |
| CN114115197A (en) * | 2021-11-24 | 2022-03-01 | 浙大城市学院 | Gas turbine state maintenance decision making system, method and storage medium |
| CN114239708B (en) * | 2021-12-13 | 2023-04-07 | 中国船舶重工集团公司第七0三研究所 | Combustion engine abnormity detection method based on quality control chart theory |
| CN114252274A (en) * | 2021-12-13 | 2022-03-29 | 中国船舶重工集团公司第七0三研究所 | Online detection method for blockage of gas inlet filter of gas turbine |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105550700A (en) * | 2015-12-08 | 2016-05-04 | 国网山东省电力公司电力科学研究院 | Time series data cleaning method based on correlation analysis and principal component analysis |
| CN105719094A (en) * | 2016-01-27 | 2016-06-29 | 刘冰 | State evaluation method of power transmission equipment |
| CN107044349A (en) * | 2017-04-17 | 2017-08-15 | 清华大学深圳研究生院 | A kind of breakdown in the motor Forecasting Methodology |
| CN108197648A (en) * | 2017-12-28 | 2018-06-22 | 华中科技大学 | A kind of Fault Diagnosis Method of Hydro-generating Unit and system based on LSTM deep learning models |
| CN108398271A (en) * | 2017-02-07 | 2018-08-14 | 山东云舜智能科技有限公司 | A kind of engine state monitor method based on vibration and nerual network technique |
| CN109086888A (en) * | 2018-09-05 | 2018-12-25 | 北京机械设备研究所 | Automobile engine failure judgment method and its device based on voice recognition |
| CN109555566A (en) * | 2018-12-20 | 2019-04-02 | 西安交通大学 | A kind of turbine rotor method for diagnosing faults based on LSTM |
-
2019
- 2019-07-04 CN CN201910601564.3A patent/CN110441065B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105550700A (en) * | 2015-12-08 | 2016-05-04 | 国网山东省电力公司电力科学研究院 | Time series data cleaning method based on correlation analysis and principal component analysis |
| CN105719094A (en) * | 2016-01-27 | 2016-06-29 | 刘冰 | State evaluation method of power transmission equipment |
| CN108398271A (en) * | 2017-02-07 | 2018-08-14 | 山东云舜智能科技有限公司 | A kind of engine state monitor method based on vibration and nerual network technique |
| CN107044349A (en) * | 2017-04-17 | 2017-08-15 | 清华大学深圳研究生院 | A kind of breakdown in the motor Forecasting Methodology |
| CN108197648A (en) * | 2017-12-28 | 2018-06-22 | 华中科技大学 | A kind of Fault Diagnosis Method of Hydro-generating Unit and system based on LSTM deep learning models |
| CN109086888A (en) * | 2018-09-05 | 2018-12-25 | 北京机械设备研究所 | Automobile engine failure judgment method and its device based on voice recognition |
| CN109555566A (en) * | 2018-12-20 | 2019-04-02 | 西安交通大学 | A kind of turbine rotor method for diagnosing faults based on LSTM |
Non-Patent Citations (1)
| Title |
|---|
| 基于ＡＬＩＦ－ＨＴ的汽轮发电机组转子故障诊断;唐贵基等;《动力工程学报》;20171130;第37卷(第11期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110441065A (en) | 2019-11-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110441065B (en) | Gas turbine on-line detection method and device based on LSTM | |
| CN111275288B (en) | XGBoost-based multidimensional data anomaly detection method and device | |
| CN110618610A (en) | Gas turbine NOx standard exceeding early warning method and device based on LSTM deep learning | |
| US11840998B2 (en) | Hydraulic turbine cavitation acoustic signal identification method based on big data machine learning | |
| CN112016251B (en) | Nuclear power device fault diagnosis method and system | |
| CN112231980B (en) | Engine life prediction method, storage medium and computing device | |
| CN109710661A (en) | Based on Global Genetic Simulated Annealing Algorithm to the method for high-pressure heater state analysis | |
| CN116383636A (en) | Coal mill fault early warning method based on PCA and LSTM fusion algorithm | |
| CN111273125A (en) | RST-CNN-based power cable channel fault diagnosis method | |
| CN113569472A (en) | Fault detection method, system, equipment and medium for liquid rocket engine | |
| Xu et al. | New RUL prediction method for rotating machinery via data feature distribution and spatial attention residual network | |
| CN111767657A (en) | Nuclear power system fault diagnosis method and system | |
| CN116821828A (en) | Multi-dimensional time sequence prediction method based on industrial data | |
| CN109522657B (en) | Gas turbine anomaly detection method based on correlation network and SVDD | |
| CN114046816B (en) | Sensor signal fault diagnosis method based on lightweight gradient lifting decision tree | |
| CN115618506A (en) | Method for predicting power of single-shaft combined cycle gas turbine | |
| CN110750756B (en) | Real-time on-line instrument checksum diagnosis method through optimal support vector machine algorithm | |
| CN110705186B (en) | Real-time online instrument checksum diagnosis method through RBF particle swarm optimization algorithm | |
| Najar et al. | Comparative Machine Learning Study for Estimating Peak Cladding Temperature in AP1000 Under LOFW | |
| CN114936520B (en) | Steam turbine fault early warning analysis method based on improved MSET | |
| CN113465930B (en) | Gas turbine multi-sensor fault detection method based on hybrid method | |
| CN117688496B (en) | Abnormality diagnosis method, system and equipment for satellite telemetry multidimensional time sequence data | |
| Chen et al. | A Multivariate Time Series Anomaly Detection Method Based on Generative Model | |
| Hu et al. | Detection of chiller energy efficiency faults using expectation maximization | |
| Gao et al. | Research on fault diagnosis of electric gate valve in nuclear power plant based on the VMD-MDI-ISSA-RF model |
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 |