CN117435936A - Cooling water temperature early warning method for water-cooled steam turbine generator - Google Patents
Cooling water temperature early warning method for water-cooled steam turbine generator Download PDFInfo
- Publication number
- CN117435936A CN117435936A CN202311356310.2A CN202311356310A CN117435936A CN 117435936 A CN117435936 A CN 117435936A CN 202311356310 A CN202311356310 A CN 202311356310A CN 117435936 A CN117435936 A CN 117435936A
- Authority
- CN
- China
- Prior art keywords
- value
- water temperature
- cooling water
- data
- window
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 47
- 239000000498 cooling water Substances 0.000 title claims abstract description 40
- 239000013598 vector Substances 0.000 claims abstract description 28
- 238000004458 analytical method Methods 0.000 claims abstract description 10
- 230000007246 mechanism Effects 0.000 claims abstract description 10
- 238000012544 monitoring process Methods 0.000 claims abstract description 6
- 238000000605 extraction Methods 0.000 claims abstract description 5
- 238000012216 screening Methods 0.000 claims abstract description 4
- 230000000007 visual effect Effects 0.000 claims abstract description 4
- 238000011176 pooling Methods 0.000 claims description 21
- 238000004364 calculation method Methods 0.000 claims description 14
- 210000004027 cell Anatomy 0.000 claims description 12
- 239000011159 matrix material Substances 0.000 claims description 8
- 230000004913 activation Effects 0.000 claims description 6
- 238000004422 calculation algorithm Methods 0.000 claims description 5
- 230000000694 effects Effects 0.000 claims description 5
- 238000012549 training Methods 0.000 claims description 5
- 238000007405 data analysis Methods 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 3
- 210000002569 neuron Anatomy 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- 230000009467 reduction Effects 0.000 claims description 3
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 abstract description 22
- 230000035945 sensitivity Effects 0.000 abstract description 2
- 238000005065 mining Methods 0.000 abstract 1
- 238000004804 winding Methods 0.000 description 14
- 230000002159 abnormal effect Effects 0.000 description 6
- 238000013527 convolutional neural network Methods 0.000 description 6
- 238000013528 artificial neural network Methods 0.000 description 5
- 230000007547 defect Effects 0.000 description 5
- 230000006870 function Effects 0.000 description 4
- 230000000875 corresponding effect Effects 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 238000005070 sampling Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 230000002596 correlated effect Effects 0.000 description 2
- 125000004122 cyclic group Chemical group 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 230000009471 action Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000001149 cognitive effect Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 239000004020 conductor Substances 0.000 description 1
- 238000001816 cooling Methods 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000002068 genetic effect Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
- 238000012800 visualization Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
- F01—MACHINES OR ENGINES IN GENERAL; ENGINE PLANTS IN GENERAL; STEAM ENGINES
- F01D—NON-POSITIVE DISPLACEMENT MACHINES OR ENGINES, e.g. STEAM TURBINES
- F01D15/00—Adaptations of machines or engines for special use; Combinations of engines with devices driven thereby
- F01D15/10—Adaptations for driving, or combinations with, electric generators
-
- 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
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/02—Total factory control, e.g. smart factories, flexible manufacturing systems [FMS] or integrated manufacturing systems [IMS]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Mechanical Engineering (AREA)
- Control Of Turbines (AREA)
Abstract
The invention provides a cooling water temperature early warning method of a water-cooled steam turbine generator. Firstly, carrying out visual analysis on DCS system acquisition variables through a thermal cluster map, screening out strong correlation characteristics which influence cooling water temperature to form a characteristic vector, inputting the characteristic vector into a CNN-GRU-ATTENTION prediction model, calculating a normal maximum residual value through a sliding window method, taking the normal maximum residual value as a fault threshold value, and then carrying out real-time monitoring and judging on the monitored water temperature residual value through the sliding window method to judge whether the monitored water temperature residual value exceeds the threshold value so as to realize water temperature overheat fault early warning. The method fully integrates the relation between hierarchical clustering heat map mining samples and the strong feature extraction capability of the CNN model, has a simple GRU network structure, is easy to converge, and has important features highlighted by an Attention mechanism, and the sensitivity of model early warning is improved by real-time monitoring by a sliding window method.
Description
Technical Field
The invention relates to the field of generator stator winding thermal defect early warning, in particular to a cooling water temperature early warning method of a water-cooled steam turbine generator.
Background
The large-sized steam turbine generator is an important electrical device of a power plant, and a thermal fault caused by blockage of a cold water system in a stator winding of the generator is one of main faults of the large-sized water-cooled steam turbine generator. The existing method for early warning the thermal faults of the stator winding of the generator is mainly based on mechanism analysis. The early research on the thermal faults of the stator winding of the generator is to analyze the blockage fault mechanism of the hollow conductor of the water-cooled stator winding bar, and the method can quantitatively calculate the standard value of the temperature of the stator winding under any working condition, and the error is within 5 ℃. However, the method has large calculated amount, can perform offline analysis, and has larger difficulty in judging abnormal working conditions in real time; some scholars establish cooling water temperature models of the stator winding under different working conditions, and respectively identify the cooling water temperature models by using a least square method and a BP neural network algorithm, wherein the error is less than 1 ℃. However, the actual working conditions are complex and changeable, the standard value cannot be applied to various working conditions, and the abnormal working conditions are difficult to accurately judge; still other students build three-dimensional models for relevant parts of the stator of the generator based on multi-field coupling analysis, perform numerical simulation, and more comprehensively describe the thermal fault process of the stator winding of the generator. The method provides a certain theoretical support for the research of the thermal faults of the stator winding of the generator, but is not suitable for online real-time evaluation and early warning.
At present, when temperature prediction research is carried out, the difference between a predicted value and an actual value of the stator core temperature of the BP neural network motor is warned to be within 3 ℃, and the combination of a support vector machine and a genetic algorithm has great advantages in model precision, optimizing training speed and generalization. However, the above traditional machine learning prediction model has higher prediction accuracy when the data volume is smaller and the input variables are single and most of non-time sequence variables, researchers usually adopt an LSTM cyclic neural network for time sequence data, the characteristic change of learning long-short-term dependent data is provided, the prediction effect of using the LSTM cyclic neural network in a time sequence prediction scene is better than that of the traditional data mining method, the LSTM is more complex in calculation compared with the simple machine learning model, and the speed of training and reasoning processes is slower when large-scale data are processed.
Disclosure of Invention
Aiming at the defects existing in the existing early warning method, the invention aims to provide the water-cooled turbine generator cooling water temperature early warning method for realizing early warning of overheat defects of the stator winding.
In order to solve the technical problems, the invention provides a cooling water temperature early warning method of a water-cooled steam turbine generator, which comprises the following steps:
step one: collecting generator operation data acquired by a generator DCS system, performing visual analysis on the monitoring data through a cluster heat map, and screening related variables which are strongly related to the cooling water temperature of the stator;
step two: inputting the screened variable into a CNN model for convolution operation, carrying out maximum value pooling operation on the data subjected to the convolution operation, and converting an input vector subjected to the convolution operation and pooling operation into an output vector;
step three: the output characteristic vector is input into a GRU network structure to carry out model training;
step four: adopting an ATTENTION mechanism, obtaining the probability corresponding to different feature vectors according to a weight distribution principle by using the feature vectors subjected to GRU network activation processing, completing a CNN-GRU-ATTENTION model, and calculating residual values of a normal working condition stator cooling water temperature predicted value and a true value;
step five: according to the obtained residual error value, setting proper window size and step length, adopting a sliding window method to conduct window sliding data analysis on the residual error value, and calculating the maximum window residual error mean value under normal working conditions;
step six: setting a threshold value through an alarm threshold value formula, calculating a window mean value through on-site data acquisition, inputting a model and a sliding window method, comparing the window mean value with the threshold value, and judging whether cooling water is in a normal temperature range or not.
In a preferred embodiment: the clustering heat map is displayed as a tree map by calculating hierarchical clustering results of Euclidean distance between data samples, and meanwhile, the Pearson correlation coefficient is calculated as the color coding of the heat map to display values in a data matrix, wherein the specific calculation formula is as follows:
the Euclidean distance formula between two variable samples is:
wherein D is xy For Euclidean distance between two samples, X n For the DCS system to collect the specific value of the variable X at the moment n, y n The DCS system is used for collecting a specific value of a variable y at a moment n,
pearson correlation coefficient formula:
wherein r is i Representing a correlation coefficient value between the i-th parameter and the target parameter; x is X ij A j-th data value representing an i-th parameter; y is j A j-th data value representing a target parameter;a data mean representing the i-th variable; />A data average representing the target parameter.
In a preferred embodiment: the CNN model is a two-layer convolution layer, each layer of 8 convolution kernels has a size of 3, and the two-layer pooling layer adopts a maximum pooling window with a step length of 2; the convolution and pooling operation of the CNN model comprises the following specific steps: characteristic variables with strong relevance with cooling water temperature are used as input variables, and representative characteristics of data are extracted through the local receptive field, so that a time sequence characteristic vector is formed; and compressing the preliminarily extracted feature vectors by a pooling window of the pooling layer to achieve a feature dimension reduction effect, and finally integrating the features in different convolution kernels and pooling layer mapping into a plurality of vectors to be input into the GRU layer to finish the extraction of the feature vectors.
In a preferred embodiment: the GRU layer is 2 layers, the number of neurons is 64 and 128, and each unit cell is provided with a refresh gate z t Reset gate r t Hidden state information selected to be left at current momentThe hidden state h of the cell output at the current moment t Each parameter expression algorithm:
the update gate expression is: z t ＝σ(W z ·[h t-1 ，x t ])
The reset gate expression is: r is (r) t ＝σ(Wr·[h t-1 ，x t )
The hidden state information expression selected and left at the current moment is as follows:
the hidden state expression of the cell output at the current moment is:
wherein z is t To update the gate calculation result value, σ is the S-type activation function, W z To reset the gate weight matrix, W r To reset the gate weight matrix, h t-1 H is the hidden state information of the last moment t The hidden state value of the cell, r, output at the current moment t Resetting the gate calculation result value, tanh is a trigonometric function, x t The characteristic information value is entered for the current time,representing the hidden state information value selected for the current time.
In a preferred embodiment: the residual error value calculation formula is as follows: r is R e =y-Y ', where Y is the true value and Y' is the predicted value.
In a preferred embodiment: the window residual error average value calculation formula is as follows:wherein R is n N is a window size, X ε Is the jth window mean.
In a preferred embodiment: the alarm threshold formula is as follows: e (E) Y ＝g 1 E V
Wherein E is Y To finally determine the threshold, g 1 Is an alarm coefficient E V Is the maximum prediction residual mean value of normal operation.
Compared with the prior art, the technical scheme of the invention has the following beneficial effects:
the invention provides a cooling water temperature early warning method of a water-cooled steam turbine generator, which uses a cluster heat map to carry out visual analysis on DCS and presents the similarity and difference among different acquisition variables in an intuitive way. This helps discover the links between the variables. The CNN-GRU-Attention model is used for combining CNN value multi-level feature extraction and GRU to capture the time dependence in data and combine the Attention mechanism to process the sequence advantages, so that the accuracy of predicted data is greatly improved. And calculating a threshold value through a sliding window method, and monitoring the working residual value of the stator cooling water in real time. The method provided by the invention monitors the fluctuation condition of the residual water temperature value in real time. By comparing the residual value in the sliding window with the normal working condition threshold value, the generator system can continuously monitor whether the cooling water temperature of the data stator is abnormal or not, and the sensitivity of overheat defect early warning of the stator winding of the generator is greatly improved.
Drawings
FIG. 1 is a flow chart of a method for early warning of overheat of cooling water temperature of a water-cooled turbo generator;
FIG. 2 is a diagram of the structure of a CNN-GRU-ATTENTION model;
FIG. 3 is a graph of analysis results of clustering heat maps of DCS system acquisition variables;
FIG. 4 is a water temperature prediction graph;
FIG. 5 is a normal condition water temperature residual error average early warning diagram;
FIG. 6 is an abnormal condition water temperature residual error mean value early warning chart.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention; it is apparent that the described embodiments are only some embodiments of the present invention, not all embodiments, and that all other embodiments obtained by persons of ordinary skill in the art without making creative efforts based on the embodiments in the present invention are within the protection scope of the present invention.
In the description of the present invention, it should be noted that the positional or positional relationship indicated by the terms such as "upper", "lower", "inner", "outer", "top/bottom", etc. are based on the positional or positional relationship shown in the drawings, are merely for convenience of describing the present invention and simplifying the description, and do not indicate or imply that the apparatus or elements referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as limiting the present invention. Furthermore, the terms "first," "second," and the like, are used for descriptive purposes only and are not to be construed as indicating or implying relative importance.
In the description of the present invention, unless explicitly specified and limited otherwise, the terms "mounted," configured to, "" engaged with, "" connected to, "and the like are to be construed broadly, and may be, for example," connected to, "wall-mounted," connected to, removably connected to, or integrally connected to, mechanically connected to, electrically connected to, directly connected to, or indirectly connected to, through an intermediary, and may be in communication with each other between two elements, as will be apparent to those of ordinary skill in the art, in view of the detailed description of the terms herein.
Referring to fig. 1 to 6, in the cooling water temperature overheat pre-warning method for a water-cooled turbo generator according to the embodiment of the present invention, firstly, a variable acquired by a DCS system is visually analyzed through a thermal cluster map, a feature vector formed by a strong correlation feature affecting the cooling water temperature is screened out, and is input into a CNN-GRU-atention prediction model, then, a normal maximum residual value is calculated through a sliding window method, and is used as a fault threshold, and then, the monitored water temperature residual value is monitored and judged in real time through the sliding window method to determine whether the monitored water temperature residual value exceeds the threshold, and the specific steps are as follows:
step one: firstly, collecting operation monitoring data through a DCS system, screening variables related to stator cooling water intensity through a clustering heat map, and calculating Euclidean distance with minimum distance as input of a mixed model to be combined into a new class until all samples are combined into a class, or a certain termination condition is met, and ending clustering.
The Euclidean distance formula is as follows:
wherein Dxy is the Euclidean distance between two samples, xn is the specific value of the variable x acquired by the DCS system at the moment n, yn is the specific value of the variable y acquired by the DCS system at the moment n;
secondly, calculating the pearson correlation coefficient among 18 different acquisition variables to be used as the basis of the color coding of the heat map, wherein the pearson correlation coefficient has the following formula:
wherein ri represents a correlation coefficient value between the i-th parameter and the target parameter; xij represents the jth data value of the ith argument; yj represents the jth data value of the target parameter; a data mean representing the i-th variable; y represents the data mean of the target parameter.
Step two: and acquiring a variable clustering heat map analysis result through a DCS system, determining the number of the clustered classes, and selecting the characteristic variable correlated with the strong target variable from the similar classes as the CNN-GRU-ATTENTION model input variable.
Next, as shown in the model framework of fig. 2, a corresponding CNN-GRU-atention neural network prediction model is built, the CNN model is set to be two layers of convolution layers, each layer of 8 convolution kernels has a size of 3, and the two layers of pooling layers select a maximum pooling window and have a step length of 2.
The convolution and pooling operation of the CNN model comprises the following specific steps:
the characteristic variable correlated with the cooling water temperature is used as an input variable to form a time-series characteristic vector by extracting the characteristic of the data. And compressing the preliminarily extracted feature vector by a pooling window of the pooling layer to achieve the feature dimension reduction effect.
Step three: integrating the features in the mapping of different convolution kernels and pooled layers into a plurality of vectors, and inputting the vectors into the GRU layer to finish the extraction of the feature vectors.
By setting the GRU layer number as two layers, the number of the layer neurons is 64 and 128 respectively, and each unit cell is provided with an update gate zt, a reset gate rt, hidden state information selected and left at the current moment and hidden state ht of the cell output at the current moment, and each parameter expression algorithm.
The update gate expression is:
z t ＝σ(W z ·[h t-1 ，x t ])
the reset gate expression is:
r t ＝σ(W r ·[h t-1 ，x t ])
the hidden state information expression selected and left at the current moment is as follows:
the hidden state expression of the cell output at the current moment is:
wherein z is t To update the gate calculation result value, σ is the S-type activation function, W z To reset the gate weight matrix, W r To reset the gate weight matrix, h t-1 H is the hidden state information of the last moment t Output at the present timeIs the hidden state value of the cell, r t Resetting the gate calculation result value, tanh is a trigonometric function, x t The characteristic information value is entered for the current time,representing the hidden state information value selected for the current time.
Step four: adopting an ATTENTION mechanism, obtaining the probability corresponding to different feature vectors according to a weight distribution principle by using the feature vectors subjected to GRU network activation processing, completing a CNN-GRU-ATTENTION model, and calculating residual values of a normal working condition stator cooling water temperature predicted value and a true value;
the water temperature residual value formula of the calculation prediction model is as follows:
R e ＝Y-Y′
wherein Y is the actual value of the cooling water temperature, and Y' is the predicted value of the cooling water temperature.
It should be noted that the Attention mechanism is a mechanism for focusing on the most important part of resources to imitate cognitive Attention, and the weight of the input variable with large influence is adjusted through the Attention mechanism so as to improve efficiency.
Step five: according to the obtained residual error value, setting proper window size and step length, adopting a sliding window method to conduct window sliding data analysis on the residual error value, and calculating the maximum window residual error mean value under normal working conditions;
the formula for calculating the residual average value of the water temperature residual window is as follows:
wherein Rn is the nth residual value in a window, n is a window size, X ε Is the J-th window mean.
Step six: setting a threshold value through an alarm threshold value formula, calculating a window mean value through on-site data acquisition, inputting a model and a sliding window method, comparing the window mean value with the threshold value, and judging whether cooling water is in a normal temperature range or not. The water temperature residual error alarm threshold value formula:
E Y ＝g 1 E V wherein EY is the finalAnd determining a threshold value, wherein g1 is an alarm coefficient, and EV calculates the maximum prediction residual error mean value of normal operation.
Specifically, the invention aims at a turbo generator of QFSN-1000-2-27 model of a certain power plant, and the unit is provided with a DCS acquisition data system, wherein the sampling frequency of the system is 1 time/1 min, and the total monitored continuous variable is 108. The set DCS acquisition system is selected for experimental study according to data of 3 months in 2019 and 2 months in 2020 and 6 months.
More specifically, in this embodiment, clustering heat map visualization analysis is performed on data collected by DCS system of 3 months in 2019 of a power plant, and the result is shown in fig. 2, where when collected data are collected into four types, a variable with a correlation coefficient with stator cooling water greater than 0.9 is used as an input variable of a prediction model. 8 state variables extracted by the clustering heat map and water temperatures of water outlets at upper layers of the notches 7, 11 and 26 in the same direction are used as input variables of a CNN-GRU-Attention temperature prediction model, and the temperature of cooling water at an outlet of a wire rod at the upper layer of the notch 12 is used as an output variable. The training set of the test model is normal operation data of the generator from 2 days of 3 months to 12 days of 3 months in 2019, and 1440 groups in total; the validation set was normal operation data from 13 days of 3 months to 15 days of 3 months of 2019, totaling 1440 x 3 groups. The obtained water temperature prediction map is shown in fig. 4.
Firstly, predicting the normal temperature data of the cooling water at the upper layer bar outlet of the notch of the slot of the upper layer of the 2021 month 25 to 27 days 12 by using a trained model, setting a sliding window method to be 20 in window size, calculating the maximum average value of a predicted residual value window when the sliding step length of the window is 1, enabling an alarm threshold formula to take an alarm coefficient as 4, and enabling the water temperature alarm threshold value of the water outlet of the stator water internal cooling winding to be 2.531. The residual value change trend under the normal condition is shown in fig. 5, and it can be obtained that the predicted value and the true value of the water temperature of the water outlet of the steam turbine generator are far lower than the alarm threshold value when the steam turbine generator works normally.
Then, a period of data before abnormal alarming of the cold water temperature of the stator winding of the generator is input into a model to obtain an abnormal state residual variation trend chart, as shown in fig. 6, the cooling water temperature of the No. 12 slot of the stator winding is in 1668 sliding windows, namely, the 1688 sampling value residual value exceeds an alarm threshold value to send out an early warning signal, the 5189 sliding window of the actual generator sends out an alarm, and the 5209 sampling value sends out an alarm. The early defect early warning effect of the generator can be achieved.
The foregoing is only a preferred embodiment of the present invention, but the design concept of the present invention is not limited thereto, and any person skilled in the art will be able to make insubstantial modifications of the present invention within the scope of the present invention disclosed herein by this concept, which falls within the actions of invading the protection scope of the present invention.
Claims (7)
1. The cooling water temperature early warning method for the water-cooled steam turbine generator is characterized by comprising the following steps of:
step one: collecting generator operation data acquired by a generator DCS system, performing visual analysis on the monitoring data through a cluster heat map, and screening relevant variables of strong stator cooling water temperature;
step two: inputting the screened variable into a CNN model for convolution operation, carrying out maximum value pooling operation on the data subjected to the convolution operation, and converting an input vector subjected to the convolution operation and pooling operation into an output vector;
step three: the output characteristic vector is input into a GRU network structure to carry out model training;
step four: adopting an ATTENTION mechanism, obtaining the probability corresponding to different feature vectors according to a weight distribution principle by using the feature vectors subjected to GRU network activation processing, completing a CNN-GRU-ATTENTION model, and calculating residual values of a normal working condition stator cooling water temperature predicted value and a true value;
step five: according to the obtained residual error value, setting proper window size and step length, adopting a sliding window method to conduct window sliding data analysis on the residual error value, and calculating the maximum window residual error mean value under normal working conditions;
step six: setting an alarm threshold, calculating a window mean value by acquiring data on site, inputting a model and a sliding window method, comparing the window mean value with the threshold, and judging whether cooling water is in a normal temperature range or not.
2. The method for pre-warning the cooling water temperature of a water-cooled turbo generator according to claim 1, wherein the method comprises the following steps: according to the clustering heat map in the step one, the hierarchical clustering result of the Euclidean distance between the data samples is displayed as a tree map, and meanwhile, the pearson correlation coefficient is calculated as the color coding of the heat map to display the value in the data matrix, wherein the specific calculation formula is as follows:
the Euclidean distance formula between two variable samples is:
wherein D is xy Is the Euclidean distance between two samples, x n For the DCS system to collect the specific value of the variable x, y at the moment n n The DCS system is used for collecting a specific value of a variable y at a moment n,
pearson correlation coefficient formula:
wherein r is i Representing a correlation coefficient value between the i-th parameter and the target parameter; x is x ij A j-th data value representing an i-th parameter; y is j A j-th data value representing a target parameter;a data mean representing the i-th variable; />A data average representing the target parameter.
3. The method for pre-warning the cooling water temperature of a water-cooled turbo generator according to claim 1, wherein the method comprises the following steps: according to the CNN model in the second step, two layers of convolution layers are adopted, the size of 8 convolution kernels in each layer is 3, the two layers of pooling layers adopt the largest pooling window, and the step length is 2; the convolution and pooling operation of the CNN model comprises the following specific steps: characteristic variables with strong relevance with cooling water temperature are used as input variables, and representative characteristics of data are extracted through the local receptive field, so that a time sequence characteristic vector is formed; and compressing the preliminarily extracted feature vectors by a pooling window of the pooling layer to achieve a feature dimension reduction effect, and finally integrating the features in different convolution kernels and pooling layer mapping into a plurality of vectors to be input into the GRU layer to finish the extraction of the feature vectors.
4. The method for pre-warning the cooling water temperature of a water-cooled turbo generator according to claim 1, wherein the method comprises the following steps: according to the GRU layer of step three, which is 2 layers, the number of neurons is 64 and 128, and each unit cell has a refresh gate z t Reset gate r t Hidden state information selected to be left at current momentThe hidden state h of the cell output at the current moment t Each parameter expression algorithm:
the update gate expression is:
z t ＝σ(W z ·[h t-1 ，x t ])
the reset gate expression is:
r t ＝σ(W r ·[h t-1 ，x t ])
the hidden state information expression selected and left at the current moment is as follows:
the hidden state expression of the cell output at the current moment is:
wherein z is t To update the gate calculation result value, σ is the S-type activation function, W z To reset the gate weight matrix, W r To reset the gate weight matrix, h t-1 H is the hidden state information of the last moment t The hidden state value of the cell, r, output at the current moment t Resetting the gate calculation result value, tanh is a trigonometric function, x t The characteristic information value is entered for the current time,representing the hidden state information value selected for the current time.
5. The method for pre-warning the cooling water temperature of a water-cooled turbo generator according to claim 1, wherein the method comprises the following steps: the residual error value calculation formula according to the fourth step is: r is R e ＝Y-Y′
Wherein Y is a true value and Y' is a predicted value.
6. The method for pre-warning the cooling water temperature of a water-cooled turbo generator according to claim 1, wherein the method comprises the following steps: the window residual error mean value calculation formula according to the fifth step is as follows:
wherein R is n N is a window size, X ε Is the jth window mean.
7. The method for pre-warning the cooling water temperature of a water-cooled turbo generator according to claim 1, wherein the method comprises the following steps: the alarm threshold formula according to the step six is: e (E) Y ＝g 1 E V
Wherein E is Y To finally determine the threshold, g 1 Is an alarm coefficient E V Is the maximum prediction residual mean value of normal operation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311356310.2A CN117435936A (en) | 2023-10-19 | 2023-10-19 | Cooling water temperature early warning method for water-cooled steam turbine generator |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311356310.2A CN117435936A (en) | 2023-10-19 | 2023-10-19 | Cooling water temperature early warning method for water-cooled steam turbine generator |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117435936A true CN117435936A (en) | 2024-01-23 |
Family
ID=89545461
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311356310.2A Pending CN117435936A (en) | 2023-10-19 | 2023-10-19 | Cooling water temperature early warning method for water-cooled steam turbine generator |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117435936A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117992762A (en) * | 2024-04-07 | 2024-05-07 | 华侨大学 | Overheat early warning method and device for stator winding of water-cooled steam turbine generator |
-
2023
- 2023-10-19 CN CN202311356310.2A patent/CN117435936A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117992762A (en) * | 2024-04-07 | 2024-05-07 | 华侨大学 | Overheat early warning method and device for stator winding of water-cooled steam turbine generator |
| CN117992762B (en) * | 2024-04-07 | 2024-07-02 | 华侨大学 | Overheat early warning method and device for stator winding of water-cooled steam turbine generator |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111596604B (en) | Intelligent fault diagnosis and self-healing control system and method for engineering equipment based on digital twinning | |
| CN110441065B (en) | Gas turbine on-line detection method and device based on LSTM | |
| CN113156917B (en) | Power grid equipment fault diagnosis method and system based on artificial intelligence | |
| CN111861274A (en) | Water environment risk prediction and early warning method | |
| CN110598736A (en) | Power equipment infrared image fault positioning, identifying and predicting method | |
| CN117435936A (en) | Cooling water temperature early warning method for water-cooled steam turbine generator | |
| CN113807951B (en) | Transaction data trend prediction method and system based on deep learning | |
| CN117726181A (en) | Collaborative fusion and hierarchical prediction method for typical disaster risk heterogeneous information of coal mine | |
| CN114676742A (en) | Power grid abnormal electricity utilization detection method based on attention mechanism and residual error network | |
| CN111273125A (en) | RST-CNN-based power cable channel fault diagnosis method | |
| CN116610998A (en) | Switch cabinet fault diagnosis method and system based on multi-mode data fusion | |
| CN111222798A (en) | Soft measurement method for key indexes of complex industrial process | |
| CN115470850A (en) | Water quality abnormal event recognition early warning method based on pipe network water quality time-space data | |
| CN115096627A (en) | Method and system for fault diagnosis and operation and maintenance in manufacturing process of hydraulic forming intelligent equipment | |
| CN116109039A (en) | Data-driven anomaly detection and early warning system | |
| Zhang et al. | Grouped time series networks for grade monitoring of zinc tailings with multisource features | |
| CN117371207A (en) | Extra-high voltage converter valve state evaluation method, medium and system | |
| CN115457307A (en) | Chemical process fault diagnosis method based on improved residual error network | |
| CN117033923A (en) | Method and system for predicting crime quantity based on interpretable machine learning | |
| CN115099332A (en) | Radar data real-time processing method and system based on mixed forest | |
| Liu et al. | Prediction of hydraulic pumps remaining useful life based on LSTM and Transform with dual self-attention | |
| CN114580798B (en) | Device point location prediction method and system based on transformer | |
| CN114004360B (en) | Intelligent equipment diagnosis equipment and method based on fuzzy expert model | |
| CN116738296B (en) | Comprehensive intelligent monitoring system for machine room conditions | |
| CN112180893B (en) | Construction method of fault-related distributed orthogonal neighborhood preserving embedded model in CSTR process and fault monitoring method thereof |
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 |