CN114004059B - Health portrait method for hydroelectric generating set - Google Patents
Health portrait method for hydroelectric generating set Download PDFInfo
- Publication number
- CN114004059B CN114004059B CN202111123896.9A CN202111123896A CN114004059B CN 114004059 B CN114004059 B CN 114004059B CN 202111123896 A CN202111123896 A CN 202111123896A CN 114004059 B CN114004059 B CN 114004059B
- Authority
- CN
- China
- Prior art keywords
- unit
- data
- tested
- dimension
- dimensional
- 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
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/2433—Single-class perspective, e.g. one-against-all classification; Novelty detection; Outlier detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/04—Ageing analysis or optimisation against ageing
-
- 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
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E10/00—Energy generation through renewable energy sources
- Y02E10/20—Hydro energy
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Biology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
- Control Of Eletrric Generators (AREA)
Abstract
The application belongs to the technical field of state monitoring of water turbine generator sets, and particularly relates to a health portrait method of a generator set. The method comprises the steps of firstly, acquiring various kinds of relevant data of a unit to be tested, combining and processing different kinds of relevant data to obtain various kinds of dimensional data; obtaining a weight coefficient corresponding to each dimension data in the multiple dimension data by using a weight evaluation model; and determining a health radar map of the key components in the unit to be tested according to the dimension categories and the corresponding weight coefficients of the multiple kinds of dimension data. The method solves the problems that the unit health assessment is partially and wholly not considered, the sensor information and the component or unit state are not really bound, the health index is single, the display mode is single and the like in the current unit state monitoring system by utilizing the artificial intelligence big data mining and the multidimensional information visualization technology.
Description
Technical Field
The application belongs to the technical field of state monitoring of water turbine generator sets, and particularly relates to a method for health portrait of a water turbine generator set.
Background
The operation and maintenance of the water turbine generator set adopt a planned maintenance scheme, and the actual operation state of the machine is not considered in the planned maintenance, so that the problem that the generator set in a perfect state is stopped for maintenance (over maintenance) and the machine at a fault edge is ignored (under maintenance) exists. With the development of information technology and artificial intelligence technology, predictive maintenance schemes based on operation state online monitoring and artificial intelligence fault diagnosis methods are researched, popularized and applied, but the artificial intelligence fault diagnosis schemes face many problems when applied to hydroelectric power generation scenes.
The single water turbine generator set mainly comprises components such as a water turbine, a generator, a technical water supply pump, a leakage drainage pump and the like, different components can operate in cooperation with the generator set to run healthily, common running state monitoring is only performed on one component, and the whole health of the display unit cannot be evaluated; even some state monitoring is only the area near the installation position of the sensor, and cannot be really bound with the unit or key components in the unit to be tested, and finally, only a fuzzy overall evaluation of the unit state in the area can be given.
Disclosure of Invention
The fault identification method aims at solving the technical problems to at least a certain extent, and based on the attention of the inventor, as the fault sample of a single water-turbine generator set is extremely scarce, a fault identification model based on supervised learning cannot be established or the performance of the model is unstable; the abnormal detection method of unsupervised learning is utilized, and usually, binary judgment on whether the abnormal condition exists can be given only according to model scoring and a preset threshold value, so that richer and more intuitive unit health information which can be used for assisting operation and maintenance decisions cannot be provided for unit operation and maintenance personnel.
In view of the above, the present disclosure provides a unit health representation method, which is used for solving the problems in the prior art that the unit health assessment is not considered locally or wholly, sensor information and component or unit states are not really bound, a health index is single, a display mode is single, and the like.
The unit health image method provided by the embodiment of the disclosure comprises the following steps:
step 1, acquiring various relevant data of a generator set to be tested, combining and processing different relevant data to obtain various dimensional data;
step 2, obtaining a weight coefficient corresponding to each dimension data in the multiple dimension data by using a weight evaluation model; and determining a healthy radar chart of key components in the generator set to be tested according to the dimension categories and the corresponding weight coefficients of the multiple kinds of dimension data.
Optionally, the multiple kinds of relevant data of the generator set to be tested include:
(a) acquiring machine set ledger data;
(b) unit installation environment data;
(c) the unit maintenance records data;
(d) unit state monitoring data;
(e) unit generation control parameters and the like;
(f) and generating performance data of the unit.
Optionally, in the step 1, different pieces of relevant data are combined and processed to obtain multiple pieces of dimensional data, where the dimensional data is an indication value of a bathtub curve dimension in the multiple pieces of dimensional data, and the method includes:
(1) respectively obtaining the fault and the service life of products of different manufacturers of each key component in different environments from machine set account data, machine set installation environment data and machine set maintenance record data, making a failure probability bathtub curve by using the following function expression, obtaining the service life stage of the key component in different to-be-tested machines by using the failure probability bathtub curve, and identifying the key component in the key to-be-tested machine set or rendering a building information model of the machine set by using different colors; key components comprise a water turbine, a generator, a technical water supply pump and a workshop seepage drainage pump;
(2) repeating the step (1), traversing all key components in the unit to be tested, identifying the current service life stage of each key component by using different colors, and rendering the building information model of the unit by using colors;
(3) and obtaining a bathtub curve of the unit according to the bathtub service life of the key component, and further calculating to obtain an indicated value of the bathtub curve dimension.
Optionally, in step 1, different relevant data are combined and processed to obtain multiple kinds of dimensional data, where the dimensional data is an indication value of an anomaly detection dimension in the multiple kinds of dimensional data, and the method includes:
(1) collecting running state monitoring data of a unit to be tested in a normal running state;
(2) preprocessing the state monitoring data by adopting a feature extraction method to obtain a sample Vector with a fixed length;
(3) repeating the step (1) and the step (2) to obtain a plurality of sample vectors, wherein the plurality of sample vectors form a normal data training sample set;
(4) inputting the normal data training sample set into an auto-supervised learning deep neural network, and training the auto-supervised learning deep neural network to obtain a deep neural network Model;
(5) inputting the high-dimensional sample Vector into the deep neural network Model in the step (4), outputting to obtain a low-dimensional feature Vector Feats, and establishing a Gaussian mixture Model GMM according to the feature Vector Feats;
(6) preprocessing the running state monitoring data Vector of the unit to be detected by adopting a feature extraction method to obtain a high-dimensional Vector;
(7) inputting the high-dimensional Vector into the deep neural network Model in the step (4), and calculating to obtain a low-dimensional feature Vector Feats;
(8) inputting the characteristic vector Fc into the Gaussian mixture model GMM in the step (5) to obtain an abnormal scoring score of the unit to be tested;
(9) and according to the calculated abnormal score of the key components in the unit to be detected, rendering a building information model of the unit by using colors, displaying historical health degree data by using a curve graph, and displaying an abnormal indication value of an abnormal detection dimension.
Optionally, in step 1, different pieces of relevant data are combined and processed to obtain multiple pieces of dimensional data, where the dimensional data is a fault indication value of a fault dimension, and the method includes:
(1) processing state monitoring data DataFault of a unit fault time period by adopting a feature extraction method to obtain a sample vector VectorFault with a fixed length;
(2) inputting the VectorFault into an automatic-supervision learning deep neural network, and training the automatic-supervision learning deep neural network to obtain a deep neural network model Fault;
(3) inputting the test vectorFault of the unit to be tested into the model Fault of the deep neural network, and outputting to obtain a low-dimensional feature vector FeatsFault;
(4) establishing a Gaussian mixture model GMM _ Fault according to the feature vector Feats;
(5) processing running state monitoring data DcFault of a unit to be detected by adopting a feature extraction method to obtain high-dimensional VcFault;
(6) inputting the high-dimensional VcFault in the step (5) into the deep neural network model ModelFault in the step (2), and outputting to obtain a low-dimensional feature vector FcFault;
(7) inputting the feature vector FcFault in the step (6) into the Gaussian mixture model GMM _ Fault in the step (4), and outputting an abnormal scoring score of the data to be detected;
(8) and rendering the building information model of the unit by using the calculated health degree index of the key component in the unit to be tested by using colors, displaying historical health degree data by using a curve diagram, and displaying a fault indication value of a fault dimension.
Optionally, combining and processing different related data to obtain multiple kinds of dimensional data, where the dimensional data is an expert score indication value of an expert dimension, and includes:
(1) analyzing the characteristic quantity of a key component in the unit to be tested by utilizing a computer expert mechanism, wherein the characteristic quantity comprises the intensity, peak-to-peak value, variance, skewness, kurtosis, sound frequency spectrum and sound envelope spectrum of a sound sensor signal, and the effective value, peak-to-peak value, variance, skewness, kurtosis, vibration frequency spectrum, vibration envelope spectrum, pendulum intensity, axis locus and short-time Fourier spectrum of a vibration sensor signal;
(2) displaying the intensity, peak-to-peak value, variance, skewness and kurtosis of the sound sensor signal and the time series curve of the effective value, peak-to-peak value, variance, skewness, kurtosis and swing intensity of the vibration sensor signal by using a curve graph;
(3) utilizing a curve graph to display a sound frequency spectrum, a sound envelope spectrum, a vibration frequency spectrum, a vibration envelope spectrum, a pendulum intensity and an axis track curve;
(4) displaying a short-time Fourier spectrum of the vibration sensor signal and the sound sensor signal by using a thermodynamic diagram;
(5) and rendering a Building Information Model (BIM) of the unit by using a curve chart and a thermodynamic diagram in colors, displaying historical health degree data by using the curve chart, and displaying an expert grading indicating value of an expert dimension.
Optionally, in step 1, different relevant data are combined and processed to obtain multiple kinds of dimensional data, where the dimensional data is an indication value of a production dimension of the unit to be tested, and includes:
(1) acquiring unit generation control parameters including parameters for controlling unit operation conditions and production performance data through manual setting;
(2) classifying the energy consumption by calculating the time power entropy, the water speed power responsivity, the water energy conversion rate, the power curve accessibility, the time availability rate and the energy availability rate of the unit;
(3) identifying the time power entropy, the water speed power responsivity, the water energy conversion rate, the power curve accessibility, the time availability rate and the energy availability rate of the unit by using different colors, rendering a building information model of the unit, and displaying a historical curve of the yield, the rejection rate and the energy consumption by using a curve graph;
(4) and identifying the load percentage of the production performance level of the key components in the unit to be tested by using different colors, rendering a building information model of the unit, and obtaining an expert score indicating value of the expert dimensionality by the load percentage of 100.
Optionally, in the step (2), a weight evaluation model is used to obtain a weight coefficient corresponding to each dimensional data in the multiple dimensional data, so as to obtain a health degree radar chart of the key component in the unit to be tested. The method comprises the following steps:
(1) carrying out weighted average on the scores of multiple dimensions of the unit to be tested to obtain the score Sc of the overall health degree of the key component in the unit to be tested;
the method comprises the following steps of A, determining the value of Sc, N and omega, wherein Sc is the overall health degree value of a key component in a unit to be tested, N is a dimension number, omega is the weight of each dimension, and the value of omega is determined according to a specific unit;
(2) and obtaining a health radar map of the key components in the unit to be tested according to the overall health score Sc.
The unit health imaging method provided by the embodiment of the disclosure solves the problems that the unit health assessment is partially and wholly not considered, the sensor information and the component or the unit state are not really bound, the health index is single, and the display mode is single in the current unit state monitoring system by utilizing the artificial intelligence big data mining and the multidimensional information visualization technology. The method comprises the steps of obtaining machine set account data, machine set installation environment data, machine set overhaul record data, machine set state monitoring data, machine set generation control parameters and machine set power generation performance data. Extracting fine characteristics related to the health degree of each component of the unit from the unit state monitoring data by using a deep neural network; analyzing and extracting data indexes related to the health degree of each component of the unit, which are contained in the unit state monitoring data, by utilizing a physical mechanism; and establishing a residual life model of each component of the unit by using a statistical learning method. Displaying the residual service life of each component by using a computer visualization technology in a scatter diagram, a curve graph, a histogram, a thermodynamic diagram and other modes; and in combination with BIM, the components and the whole machine set of the three-dimensional model can be rendered in different colors at different health levels for three-dimensional visual display.
Additional aspects and advantages of the disclosure will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the invention.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below. It is obvious that the drawings in the following description are only some embodiments of the invention, and that for a person skilled in the art, other drawings can be derived from them without inventive effort.
FIG. 1 is a schematic diagram illustrating a probability of failure bathtub curve according to one embodiment of the present disclosure.
Fig. 2 is a health radar chart shown according to one embodiment of the present disclosure.
Detailed Description
The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
The unit health image method provided by one embodiment of the disclosure can include the following steps:
in step 1, various kinds of relevant data of the unit to be tested are obtained, different kinds of relevant data are combined and processed, and various kinds of dimensional data are obtained.
In one embodiment, the plurality of kinds of relevant data of the unit under test include:
(a) acquiring machine set account data (ED1) including manufacturers, delivery batches, online service time and the like of machines and components;
(b) plant installation environmental data (ED2) including temperature, humidity, etc. of the plant installation area;
(c) the unit maintenance record data (ED3) comprises unit inspection records and all levels of maintenance records, specifically comprises fault discovery time, maintenance content, component replacement, wear degree measurement of worn components and the like;
(d) the unit state monitoring data (ED4) comprises data recorded by vibration, temperature, pressure, current, voltage and other monitoring sensors;
(e) the unit generation control parameters (ED5) comprise parameters which can be manually set to control the operation conditions of the unit, such as exciting current, guide vane opening and the like;
(f) the unit power generation performance data (ED6) comprises power generation time power entropy, water speed power responsivity, water energy conversion rate, power curve accessibility, time availability rate, energy availability rate and the like.
In one embodiment, the dimensional data is an indicated value of a bathtub curve dimension in multiple dimensional data, data suitable for manufacturing the bathtub curve dimension is obtained by combining original data of a unit to be tested, the indicated value of the bathtub curve dimension is obtained, and a building information model (BIM is based on a three-dimensional digital technology, integrates an engineering data model of various relevant information of a building engineering project, is a detailed expression of the relevant information of the engineering project, and includes:
(1) respectively obtaining the faults and the service lives of products of different manufacturers of each key component in different environments from machine group account data (ED1), machine group installation environment data (ED2) and machine group overhaul record data (ED3), making failure probability bathtub curves by using the following function expressions, obtaining the service life stages of the key components in different machine groups to be tested by using the failure probability bathtub curves, and identifying the key components in the key machine groups to be tested by using different colors or rendering Building Information Models (BIM) of the machine groups;
the bathtub curve function expression is:
λ(t)＝γβ(t/α) β-1 exp((t/α) β )，
wherein: alpha, beta and gamma are parameters in a distribution function, and the numerical type is a rational score;
(2) repeating the step (1), traversing all key components in the unit to be tested, identifying the current service life stage of each key component by using different colors, and rendering a Building Information Model (BIM) of the unit by using colors;
(3) obtaining a bathtub curve of the unit according to the bathtub service life of the key component, and further calculating to obtain an indicated value of the bathtub curve dimension:
Sc1＝a*|t-t total /2|*100，
where t is the service date of the key component, t total Is the service life of the key component, a is the adjustment coefficient, and the value range of a is [0,1 ]]。
In one embodiment, in the step 1, different pieces of relevant data are combined and processed to obtain a plurality of kinds of dimensional data, where the dimensional data is an indication value of an anomaly detection dimension in the plurality of kinds of dimensional data, and the method includes:
establishing a normal state model for key components in the unit to be tested without fault samples by adopting full normal data, and performing health assessment by using an anomaly monitoring method;
(1) collecting running state monitoring data (ED4) of the unit to be tested in a normal running state;
(2) preprocessing the state monitoring data (ED4) by adopting a feature extraction method to obtain a sample Vector with a fixed length;
(3) repeating the step (1) and the step (2) to obtain a plurality of sample vectors, wherein the plurality of sample vectors form a normal data training sample set;
(4) inputting the normal data training sample set into an auto-supervised learning deep neural network, and training the auto-supervised learning deep neural network to obtain a deep neural network Model;
after a large amount of normally-running state data (ED4) are processed, vectors in a training sample set are input into a self-supervised learning deep neural network, such as a stacked self-encoder model and a deep confidence network model, for training; obtaining a deep neural network Model through self-supervision learning training; the model can reduce the Vector of high dimension to a feature Vector Feats of fixed length of low dimension;
(5) inputting the high-dimensional sample Vector into the deep neural network Model in the step (4), outputting to obtain a low-dimensional feature Vector Feats, and establishing a Gaussian mixture Model GMM according to the feature Vector Feats;
(6) preprocessing the running state monitoring data Vector of the unit to be detected by adopting a feature extraction method to obtain a high-dimensional Vector;
(7) inputting the high-dimensional Vector into the deep neural network Model in the step (4), and calculating to obtain a low-dimensional feature Vector Feats;
(8) inputting the characteristic vector Fc into the Gaussian mixture model GMM in the step (5) to obtain an abnormal score of the unit to be tested; the higher the score is, the higher the similarity between the data to be tested and the data in the training data set is, the more normal the state is, the higher the corresponding health degree is; and vice versa.
(9) And according to the calculated abnormal score of the key components in the unit to be detected, rendering a Building Information Model (BIM) of the unit by using colors, displaying historical health degree data by using a curve graph, and displaying an abnormal indication value of an abnormal detection dimension.
In one embodiment, the dimension data is a fault indication value of the fault dimension, and a fault fingerprint of the unit is established by using a fault sample of the unit to be tested to obtain the fault indication value of the fault dimension. The method comprises the following steps:
establishing a fault fingerprint of a fault of a unit for the key component fault type in the unit to be tested, which has accumulated a certain data fault sample, for identifying and predicting the fault of the unit;
(1) processing state monitoring data DataFault of a unit fault time period by adopting a feature extraction method to obtain a sample vector VectorFault with a fixed length;
(2) inputting the vectorFault into an auto-supervised learning deep neural network, such as a stacked auto-encoder model, a deep confidence network model and the like, and training the auto-supervised learning deep neural network to obtain a deep neural network model Fault;
inputting the VectorFault in the training sample set into a deep neural network for self-supervision learning, such as a stacked self-encoder model and a deep confidence network model for training; obtaining a deep neural network model ModelFault through self-supervision learning training; the model can reduce the dimension of the high-dimensional vector to a low-dimensional fixed-length feature vector FeatsFault;
(3) inputting the test vectorFault of the unit to be tested into the model Fault of the deep neural network, and outputting to obtain a low-dimensional feature vector FeatsFault;
(4) establishing a Gaussian mixture model GMM _ Fault according to the feature vector Feats;
(5) processing running state monitoring data DcFault of a unit to be detected by adopting a feature extraction method to obtain high-dimensional VcFault;
(6) inputting the high-dimensional VcFault in the step (5) into the deep neural network model ModelFault in the step (2), and outputting to obtain a low-dimensional feature vector FcFault;
(7) inputting the feature vector FcFault in the step (6) into the Gaussian mixture model GMM _ Fault in the step (4), and outputting an abnormal score of the data to be tested, wherein the higher the score is, the higher the similarity between the data to be tested and the data in the training data set is, the more normal the state is, and the higher the corresponding health degree is; and vice versa.
And (4) rendering the Building Information Model (BIM) of the unit by using the health degree index of the key component in the unit to be tested, which is obtained by the calculation in the step (4.7), displaying historical health degree data by using a curve diagram, and displaying a fault indication value of a fault dimension.
In one embodiment, the dimension data is expert scoring indicated values of expert dimensions, fault monitoring features are manufactured by utilizing an expert analysis mechanism, expert scoring indicated values of the expert dimensions are obtained, and forms and time-varying trends of display features such as a curve graph and a thermodynamic diagram are obtained. The method comprises the following steps:
(1) analyzing the characteristic quantity of a key component in the unit to be tested by utilizing a computing expert mechanism, wherein the characteristic quantity comprises the intensity, the peak-peak value, the variance, the skewness, the kurtosis, the sound frequency spectrum and the sound envelope spectrum of a sound sensor signal, and the effective value, the peak-peak value, the variance, the skewness, the kurtosis, the vibration frequency spectrum, the vibration envelope spectrum, the pendulum intensity, the axis locus and the short-time Fourier spectrum of a vibration sensor signal;
(2) displaying the intensity, peak-to-peak value, variance, skewness and kurtosis of the sound sensor signal and the time series curve of the effective value, peak-to-peak value, variance, skewness, kurtosis and swing intensity of the vibration sensor signal by using a curve graph;
(3) utilizing a curve graph to display a sound frequency spectrum, a sound envelope spectrum, a vibration frequency spectrum, a vibration envelope spectrum, a swing intensity and an axis track curve;
(4) displaying a short-time Fourier spectrum of the vibration sensor signal and the sound sensor signal by using a thermodynamic diagram;
(5) and rendering a Building Information Model (BIM) of the unit by using a curve chart and a thermodynamic diagram in colors, displaying historical health degree data by using the curve chart, and displaying an expert grading indicating value of an expert dimension.
In one embodiment, the dimension data is an indication value of a production dimension of the unit to be tested, and includes:
(1) acquiring unit power generation control parameters (ED5), including parameters for controlling the unit operation conditions, such as exciting current, guide vane opening and the like, through manual setting; production performance data (ED6) including power generation time power entropy, water velocity power responsivity, water energy conversion, power curve accessibility, time availability and energy availability;
(2) and calculating power generation time power entropy, water speed power responsivity, water energy conversion rate, power curve accessibility, time availability rate and energy availability rate, and grading the generation performance of the unit.
(3) Identifying the time power entropy, the water speed power responsivity, the water energy conversion rate, the power curve accessibility, the time availability rate and the energy availability rate of the unit by using different colors, rendering a Building Information Model (BIM) of the unit, and displaying a historical curve of yield, rejection rate and energy consumption by using a curve graph;
(4) and identifying the load percentage of the production performance level of the key components in the unit to be tested by using different colors, and rendering a Building Information Model (BIM) of the unit. The load percentage 100 is used to obtain the expert score indication value of the expert dimension.
In the step 2, a weight evaluation model is utilized to obtain a weight coefficient corresponding to each dimensional data in the multiple dimensional data, and the scores of multiple dimensions of the unit to be tested are weighted and averaged to obtain a total health score Sc of key components in the unit to be tested;
wherein, Sc is the total health degree score of key components in the unit to be tested, N is the dimension number, Sc 1 -Sc N The values of bathtub indication, abnormal indication, fault indication, expert scoring, production performance and the like are respectively given, omega is the weight of each dimension, and the value of omega is determined according to a specific unit.
And obtaining a health degree radar chart of key components in the unit to be tested according to the total health degree score Sc.
According to one embodiment of the present disclosure, the impact of 5 dimensions (or sides) on the overall health of the unit, such as bathtub indication, anomaly indication, fault indication, expert scoring, and production performance, is derived. The health radar map is shown in fig. 2, and the smaller the value of each dimension on the health radar map, that is, the closer to the central point, the greater the negative influence of the dimension on the overall health of the unit is. The black part in the radar chart represents the health degree of the unit and represents the qualitative description of the health degree of the unit. The larger the area of the black part is, the healthier the unit is.
Compared with a single-dimensional method, the multi-dimensional health assessment direction adopted by the method is more comprehensive in evaluation, and the problems of missed report and high false report of the traditional method by using single index alarm can be effectively reduced; and the operation and maintenance supervision personnel are assisted to quickly and accurately grasp key information by adopting abundant visual technologies to display the multidimensional health information, so that the overall and local health states of the unit are comprehensively grasped, and the operation and maintenance decision is optimized.
While the foregoing is directed to the preferred embodiment of the present disclosure, it will be understood by those skilled in the art that various changes and modifications may be made without departing from the principles of the disclosure, and it is intended that such changes and modifications be covered by the appended claims.
Claims (5)
1. A method for representing health of a hydroelectric generating set is characterized by comprising the following steps:
step 1, acquiring various relevant data of a unit to be tested, combining and processing different relevant data to obtain various dimensional data, wherein the various relevant data of the unit to be tested comprise:
(a) acquiring machine set account data, including manufacturers of machines and components, delivery batches and online service time;
(b) unit installation environment data including temperature and humidity of a unit installation area;
(c) the unit maintenance record data comprises unit inspection records and all levels of maintenance records, specifically fault finding time, maintenance content, component replacement and wear degree measurement of worn components;
(d) the unit state monitoring data comprises data recorded by vibration, temperature, pressure, current and voltage monitoring sensors;
(e) the method comprises the following steps that a unit generates control parameters, including parameters for controlling the operation condition of the unit through manual setting, specifically exciting current and guide vane opening;
(f) generating performance data of the unit comprises generating time power entropy, water speed power responsivity, water energy conversion rate, power curve accessibility, time availability rate and energy availability rate;
step 2, obtaining a weight coefficient corresponding to each dimension data in the multiple dimension data by using a weight evaluation model, and obtaining a health degree radar chart of a key component in the unit to be tested, wherein the specific process is as follows:
(1) carrying out weighted average on the scores of multiple dimensions of the unit to be tested to obtain the score Sc of the overall health degree of the key component in the unit to be tested;
wherein Sc is the total health degree score of key components in the unit to be tested, N is the dimensionality, omega is the weight of each dimensionality, and the value of omega is determined according to a specific unit;
(2) obtaining a health radar map of key components in the unit to be tested according to the overall health score Sc;
wherein, the combining and processing different related data to obtain multiple dimension data, the dimension data being an indicated value of bathtub curve dimension in the multiple dimension data, includes:
(1) respectively obtaining the fault and service life of products of different manufacturers of each key component in different environments from the machine set account data, the machine set installation environment data and the machine set maintenance record data, making a failure probability bathtub curve by using the following function expression, obtaining the service life stage of the key component in different to-be-tested machine sets by using the failure probability bathtub curve, identifying the key component in the key to-be-tested machine set by using different colors or rendering a building information model of the machine set,
the bathtub curve function expression is:
λ(t)＝γβ(t/α) β-1 exp((t/α) β )，
wherein: alpha, beta and gamma are parameters in a distribution function, and the numerical type is a rational score;
(2) repeating the step (1), traversing all key components in the unit to be tested, identifying the current service life stage of each key component by using different colors, and rendering the building information model of the unit by using colors;
(3) and obtaining a bathtub curve of the unit according to the bathtub service life of the key component, and further calculating to obtain an indicated value of the bathtub curve dimension.
2. The machine set health image method according to claim 1, wherein the different related data are combined and processed to obtain a plurality of dimensional data, wherein the dimensional data is an indication value of an anomaly detection dimension in the plurality of dimensional data, and the method comprises:
(1) collecting running state monitoring data of a unit to be tested in a normal running state;
(2) preprocessing the state monitoring data by adopting a feature extraction method to obtain a sample Vector with a fixed length;
(3) repeating the step (1) and the step (2) to obtain a plurality of sample vectors, wherein the plurality of sample vectors form a normal data training sample set;
(4) inputting the normal data training sample set into an automatic supervision learning deep neural network, and training the automatic supervision learning deep neural network to obtain a deep neural network Model;
(5) inputting the high-dimensional sample Vector into the deep neural network Model in the step (4), outputting to obtain a low-dimensional feature Vector Feats, and establishing a Gaussian mixture Model GMM according to the feature Vector Feats;
(6) preprocessing the running state monitoring data Vector of the unit to be detected by adopting a feature extraction method to obtain a high-dimensional Vector;
(7) inputting the high-dimensional Vector into the deep neural network Model in the step (4), and calculating to obtain a low-dimensional feature Vector Feats;
(8) inputting the characteristic vector Fc into the Gaussian mixture model GMM in the step (5) to obtain an abnormal score of the unit to be tested;
(9) and according to the calculated abnormal score of the key components in the unit to be detected, rendering the building information model of the unit by using colors, displaying the historical health degree data by using a curve graph, and displaying an abnormal indicated value of the abnormal detection dimension.
3. The machine set health imaging method according to claim 1, wherein the different related data are combined and processed to obtain a plurality of dimensional data, wherein the dimensional data is a fault indication value of a fault dimension, and the method comprises:
(1) processing state monitoring data DataFault of a unit fault time period by adopting a feature extraction method to obtain a sample vector VectorFault with a fixed length;
(2) inputting the VectorFault into an automatic-supervision learning deep neural network, and training the automatic-supervision learning deep neural network to obtain a deep neural network model Fault;
(3) inputting the test vectorFault of the unit to be tested into the model Fault of the deep neural network, and outputting to obtain a low-dimensional feature vector FeatsFault;
(4) establishing a Gaussian mixture model GMM _ Fault according to the feature vector Feats;
(5) processing running state monitoring data DcFault of a unit to be detected by adopting a feature extraction method to obtain high-dimensional VcFault;
(6) inputting the high-dimensional VcFault in the step (5) into the deep neural network model ModelFault in the step (2), and outputting to obtain a low-dimensional feature vector FcFault;
(7) inputting the feature vector FcFault in the step (6) into the Gaussian mixture model GMM _ Fault in the step (4), and outputting an abnormal scoring score of the data to be detected;
(8) and rendering the building information model of the unit by using the calculated health degree index of the key component in the unit to be tested by using colors, displaying historical health degree data by using a curve diagram, and displaying a fault indication value of a fault dimension.
4. The machine set health imaging method according to claim 1, wherein the different related data are combined and processed to obtain a plurality of dimensional data, wherein the dimensional data is an expert score indication value of an expert dimension, and the method comprises:
(1) analyzing the characteristic quantity of a key component in the unit to be tested by utilizing a computer expert mechanism, wherein the characteristic quantity comprises the intensity, peak-to-peak value, variance, skewness, kurtosis, sound frequency spectrum and sound envelope spectrum of a sound sensor signal, and the effective value, peak-to-peak value, variance, skewness, kurtosis, vibration frequency spectrum, vibration envelope spectrum, pendulum intensity, axis locus and short-time Fourier spectrum of a vibration sensor signal;
(2) displaying the intensity, peak-to-peak value, variance, skewness and kurtosis of the sound sensor signal and the time series curve of the effective value, peak-to-peak value, variance, skewness, kurtosis and swing intensity of the vibration sensor signal by using a curve graph;
(3) utilizing a curve graph to display a sound frequency spectrum, a sound envelope spectrum, a vibration frequency spectrum, a vibration envelope spectrum, a swing intensity and an axis track curve;
(4) displaying a short-time Fourier spectrum of the vibration sensor signal and the sound sensor signal by using a thermodynamic diagram;
(5) and rendering a Building Information Model (BIM) of the unit by using a curve chart and a thermodynamic diagram in colors, displaying historical health degree data by using the curve chart, and displaying an expert grading indicating value of an expert dimension.
5. The unit health image method according to claim 1, wherein the different related data are combined and processed to obtain a plurality of dimensional data, wherein the dimensional data is an indication value of a production dimension of the unit to be measured, and the method comprises:
(1) acquiring unit generation control parameters including parameters for controlling unit operation conditions and production performance data through manual setting;
(2) classifying the energy consumption by calculating the time power entropy, the water speed power responsivity, the water energy conversion rate, the power curve accessibility, the time availability rate and the energy availability rate of the set;
(3) identifying the time power entropy, the water speed power responsivity, the water energy conversion rate, the power curve accessibility, the time availability rate and the energy availability rate of the unit by using different colors, rendering a building information model of the unit, and displaying a historical curve of the yield, the rejection rate and the energy consumption by using a curve graph;
(4) and identifying the load percentage of the power generation performance level of the key components in the unit to be tested by using different colors, rendering a building information model of the unit, and obtaining an expert score indicating value of the expert dimension by the load percentage of 100.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111123896.9A CN114004059B (en) | 2021-09-24 | 2021-09-24 | Health portrait method for hydroelectric generating set |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111123896.9A CN114004059B (en) | 2021-09-24 | 2021-09-24 | Health portrait method for hydroelectric generating set |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114004059A CN114004059A (en) | 2022-02-01 |
| CN114004059B true CN114004059B (en) | 2022-08-23 |
Family
ID=79921910
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111123896.9A Active CN114004059B (en) | 2021-09-24 | 2021-09-24 | Health portrait method for hydroelectric generating set |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114004059B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115222164B (en) * | 2022-09-20 | 2022-12-27 | 国能大渡河大数据服务有限公司 | Water pump fault prediction method and system based on empirical coupling function |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112116262A (en) * | 2020-09-24 | 2020-12-22 | 华能盐城大丰新能源发电有限责任公司 | Evaluation method for health degree of wind generating set equipment |
| AU2021100614A4 (en) * | 2021-02-01 | 2021-04-22 | Southwest University | A novel regression prediction method for electronic nose based on broad learning system |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6686879B2 (en) * | 1998-02-12 | 2004-02-03 | Genghiscomm, Llc | Method and apparatus for transmitting and receiving signals having a carrier interferometry architecture |
| CN107341349B (en) * | 2017-06-30 | 2020-04-14 | 上海交通大学 | Method and system for evaluating health of fan, memory and controller |
| US11474893B2 (en) * | 2019-12-11 | 2022-10-18 | Qualitrol Company Llc | Method and device for assessing overall asset condition |
| CN111751117A (en) * | 2020-04-10 | 2020-10-09 | 中国人民解放军92228 部队 | Special vehicle health state assessment method and device |
| CN112653991A (en) * | 2020-12-23 | 2021-04-13 | 北京工业大学 | WLAN indoor positioning method of TebNet neural network model based on deep learning |
-
2021
- 2021-09-24 CN CN202111123896.9A patent/CN114004059B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112116262A (en) * | 2020-09-24 | 2020-12-22 | 华能盐城大丰新能源发电有限责任公司 | Evaluation method for health degree of wind generating set equipment |
| AU2021100614A4 (en) * | 2021-02-01 | 2021-04-22 | Southwest University | A novel regression prediction method for electronic nose based on broad learning system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114004059A (en) | 2022-02-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110414155B (en) | Fan component temperature abnormity detection and alarm method with single measuring point | |
| CN111539553B (en) | Wind turbine generator fault early warning method based on SVR algorithm and off-peak degree | |
| CN111444940A (en) | Fault diagnosis method for critical parts of fan | |
| CN109543210A (en) | A kind of Wind turbines failure prediction system based on machine learning algorithm platform | |
| CN113255848B (en) | Water turbine cavitation sound signal identification method based on big data learning | |
| CN107844067B (en) | A kind of gate of hydropower station on-line condition monitoring control method and monitoring system | |
| CN110405537A (en) | A kind of method for building up of the guide precision prediction model based on deep learning | |
| CN114565006A (en) | Wind driven generator blade damage detection method and system based on deep learning | |
| CN114004059B (en) | Health portrait method for hydroelectric generating set | |
| Zhu et al. | Operational state assessment of wind turbine gearbox based on long short-term memory networks and fuzzy synthesis | |
| CN115496108A (en) | Fault monitoring method and system based on manifold learning and big data analysis | |
| CN112798290B (en) | Abnormal state monitoring method of gas turbine based on spectrum reconstruction error | |
| Maurya et al. | Intelligent hybrid scheme for health monitoring of degrading rotary machines: An adaptive fuzzy c-means coupled with 1-D CNN | |
| Ding et al. | Artificial intelligence based abnormal detection system and method for wind power equipment | |
| CN112016193B (en) | Online prediction method and system for lubrication failure of shield tunneling machine system | |
| CN116538092B (en) | Compressor on-line monitoring and diagnosing method, device, equipment and storage medium | |
| CN116561691A (en) | Power plant auxiliary equipment abnormal condition detection method based on unsupervised learning mechanism | |
| CN116204825A (en) | Production line equipment fault detection method based on data driving | |
| CN115575127A (en) | Bearing residual life prediction method based on whale optimization LSSVM | |
| Souza et al. | Evaluation of data based normal behavior models for fault detection in wind turbines | |
| CN114024820B (en) | Equipment health portrait method | |
| CN115326393A (en) | Wind turbine generator bearing pair fault diagnosis method based on temperature information | |
| CN112464151A (en) | Wind turbine generator yaw system abnormal sound diagnosis method based on acoustic diagnosis | |
| Zhao et al. | Fault Diagnosis of Rolling Bearings based on GA-SVM model | |
| Zhao et al. | Refined composite hierarchical multiscale Lempel-Ziv complexity: A quantitative diagnostic method of multi-feature fusion for rotating energy devices |
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 |