CN116561691A - Power plant auxiliary equipment abnormal condition detection method based on unsupervised learning mechanism - Google Patents
Power plant auxiliary equipment abnormal condition detection method based on unsupervised learning mechanism Download PDFInfo
- Publication number
- CN116561691A CN116561691A CN202310550153.2A CN202310550153A CN116561691A CN 116561691 A CN116561691 A CN 116561691A CN 202310550153 A CN202310550153 A CN 202310550153A CN 116561691 A CN116561691 A CN 116561691A
- Authority
- CN
- China
- Prior art keywords
- data
- power plant
- auxiliary equipment
- unsupervised learning
- learning mechanism
- 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
- 230000002159 abnormal effect Effects 0.000 title claims abstract description 40
- 238000001514 detection method Methods 0.000 title claims abstract description 35
- 230000007246 mechanism Effects 0.000 title claims abstract description 20
- 238000000034 method Methods 0.000 claims abstract description 44
- 238000012549 training Methods 0.000 claims abstract description 21
- 230000005856 abnormality Effects 0.000 claims abstract description 15
- 238000011156 evaluation Methods 0.000 claims abstract description 11
- 238000012360 testing method Methods 0.000 claims abstract description 8
- 238000013507 mapping Methods 0.000 claims description 17
- 238000013528 artificial neural network Methods 0.000 claims description 16
- 238000005311 autocorrelation function Methods 0.000 claims description 10
- 230000003595 spectral effect Effects 0.000 claims description 8
- 239000012530 fluid Substances 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 4
- 238000000605 extraction Methods 0.000 claims description 4
- 230000006870 function Effects 0.000 claims description 4
- 238000012544 monitoring process Methods 0.000 claims description 4
- 238000012706 support-vector machine Methods 0.000 claims description 4
- 238000012545 processing Methods 0.000 abstract description 2
- 238000004458 analytical method Methods 0.000 description 6
- 238000010248 power generation Methods 0.000 description 6
- 230000008859 change Effects 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 4
- 230000004927 fusion Effects 0.000 description 3
- 238000001816 cooling Methods 0.000 description 2
- 230000008878 coupling Effects 0.000 description 2
- 238000010168 coupling process Methods 0.000 description 2
- 238000005859 coupling reaction Methods 0.000 description 2
- 238000000354 decomposition reaction Methods 0.000 description 2
- 238000012423 maintenance Methods 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 238000009423 ventilation Methods 0.000 description 2
- 238000005299 abrasion Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 230000002035 prolonged effect Effects 0.000 description 1
- 238000003860 storage Methods 0.000 description 1
- 239000002918 waste heat Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—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
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/10—Machine learning using kernel methods, e.g. support vector machines [SVM]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/088—Non-supervised learning, e.g. competitive learning
-
- 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
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Medical Informatics (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
The invention belongs to the field of intelligent power plant equipment application, and particularly relates to a power plant auxiliary equipment abnormal condition detection method based on an unsupervised learning mechanism, which comprises the following steps: extracting data in a normal state from a database and constructing a training data set; extracting features of the training data set and training an anomaly detection model; obtaining an evaluation value of the data set by using the anomaly detection model; and selecting a test data set, taking the test data set as the input of an abnormality detection model, and taking the evaluation values of a plurality of continuous data points as the basis for judging the abnormal working condition of the equipment. The method has the advantages of no need of manually marking data, strong self-adaptive learning capability, strong adaptability, strong robustness and capability of processing large-scale data. The method does not need to manually label the data, can automatically learn the characteristics and modes from the data, and saves a great deal of time and labor cost.
Description
Technical Field
The invention belongs to the field of intelligent power plant equipment application, and particularly relates to a power plant auxiliary equipment abnormal condition detection method based on an unsupervised learning mechanism.
Background
Auxiliary equipment of a power plant is a series of equipment arranged for assisting the operation of a generator set, and comprises a water supply system, a circulating water system, a ventilation system, a cooling system and the like. These devices do not directly participate in the power generation process, but have an important role in ensuring safe and stable operation of the generator set and improving the power generation efficiency: the water supply system and the circulating water system can effectively recover waste heat, and improve the thermal efficiency of the boiler, thereby improving the power generation efficiency of the whole power station; the ventilation system and the cooling system can control the temperature and the humidity of the generator set, and ensure the normal operation of the generator set. The auxiliary equipment of the power plant can effectively control the working environment of the generator set, reduce the loss and abrasion of the equipment and prolong the service life of the equipment. Therefore, auxiliary equipment of the power plant is an essential component part of power generation operation of the power station, and plays an important role in guaranteeing safe and stable operation of a generator set and improving power generation efficiency.
The abnormal working condition of auxiliary equipment can damage the equipment, influence the reliability and stability of the equipment, and further threaten the operation of power generation equipment and the safety of personnel. Further, this can lead to plant outages or equipment damage, which can result in significant economic losses to the plant. By detecting the abnormal working condition of auxiliary equipment, the problem of the equipment can be found in advance, the equipment can be maintained or replaced in time, hidden danger is eliminated, and the safe operation of the power system is ensured. Meanwhile, the operation condition of the equipment is known as soon as possible, so that the maintenance plan of the equipment is optimized, the service life of the equipment is prolonged, and the efficiency of the equipment is improved. In general, the method is very important for detecting the abnormal working condition of auxiliary equipment of the power plant, can ensure the safe operation of a power system, optimize the equipment maintenance plan, improve the reliability and stability of the equipment and reduce the production cost.
At present, a plurality of detection methods are available for the abnormal working conditions of auxiliary equipment of a power plant. One method aims at the driving motor, and by carrying out mechanism analysis on vibration data of the support bearing, if the vibration amplitude value or the vibration phase difference value exceeds a preset threshold value, the condition that the foundation bolt of the driving motor is possibly loosened or the angle of the broken coupling is possibly misaligned is judged; the other method is that through carrying out association relation analysis on fault data and coupling data and determining association rules, a fault prediction model is generated according to the association rules; still another method uses the difference between the predicted value of the regression model and the actual data to perform fault detection.
The prior related patent has the patent number of CN202111637049.4, and the invention patent named as a method, a system, equipment and a storage medium for identifying faults of auxiliary equipment of a thermal power plant, which comprises the following contents: the invention provides a fault identification method for auxiliary equipment of a thermal power plant, which comprises the steps of collecting auxiliary operation parameters at historical fault moments and marking data according to fault types; performing wavelet decomposition reconstruction on fault data; clustering the reconstructed historical fault data; normalizing the historical data; and classifying and matching the clustered historical fault data and the fault data to be identified based on the change value and the actual value of the operation parameters, and outputting the fault type. Said invented patent uses wavelet decomposition reconstruction to filter high-frequency noise, restore real representation of fault data, then makes clustering on the fault data, then uses two dimensionalities of change value and actual value of parameter to compare the data to be identified and historical fault data. The patent uses the marked fault data to construct a fault detection model, however, in the actual production process of the thermal power plant, the occurrence of faults is very few, and under the condition of insufficient number of fault samples, the model can be over-fitted, so that the generalization capability in the actual application is poor; meanwhile, the normal working condition data of auxiliary equipment of the thermal power plant are usually far more than the fault data, and the unbalanced data distribution can cause the model to be fitted to the normal working condition during training, so that the recognition capability of the model to the fault condition is weaker; in addition, the failure mode of a thermal power plant may change over time, and the model needs to be updated periodically with new failure data, but acquiring and labeling such data can be time consuming and costly.
Disclosure of Invention
The application aims at the problems in the prior art, and provides an abnormal working condition detection method based on normal working condition data and an unsupervised learning mechanism during the operation of auxiliary equipment of a power plant.
In order to achieve the technical effects, the technical content of the application is as follows:
an abnormal condition detection method of auxiliary equipment of a power plant based on an unsupervised learning mechanism comprises the following steps:
step (1): extracting data in a normal state from a database and constructing a training data set;
step (2): extracting features of the training data set and training an anomaly detection model;
step (3): obtaining an evaluation value of the data set by using the anomaly detection model;
step (4): and selecting a test data set, taking the test data set as the input of an abnormality detection model, and taking the evaluation values of a plurality of continuous data points as the basis for judging the abnormal working condition of the equipment.
Further, the data in the database in the step (1) includes real-time operation data and historical operation data of auxiliary equipment of the power plant.
Still further, the real-time operation data and the historical operation data of the auxiliary equipment of the power plant include: (1) basic operational data: power and rotational speed of the device; (2) monitoring data: vibration and sound generated during operation of the device; (3) indirect data: pressure, flow rate, temperature of the fluid in the conduit.
Further, the feature extracted by the feature extraction in the step (2) includes:
1) Time domain features: mean, variance, minimum, maximum, range, skewness, kurtosis, autocorrelation function, partial autocorrelation function, moving average, moving standard deviation, and exponentially weighted moving average;
2) Frequency domain characteristics: fourier transform coefficients, power spectral density, band energy ratio, frequency entropy;
3) Time-frequency domain characteristics: wavelet transform coefficients, instantaneous frequency, energy spectral density.
Further, the characteristic values are analyzed, weighted and fused, and new and more representative characteristics are formed by fusing a plurality of characteristic values, so that more information is provided, and the prediction accuracy of the model is improved.
Further, the abnormality detection model in the step (2) models data of normal working conditions in the history data obtained from the step (1) based on a One Class Deep SVDD algorithm; the specific method comprises the following steps: training a single class support vector machine model using deep neural network to sample dataFrom the input space-> Is a real set with dimension d and is mapped to the output space of high dimension +.> For a real set of dimensions p, and find a hypersphere in the space, all normal samples are contained, whose objective function is:
wherein x is i For sample 1, c and R are the center and radius of the hypersphere in the mapping space,for the data expression of the data after the data is mapped by the deep neural network f, L and W are the total layer number and weight of the neural network respectively, W l As the weight of the first layer, II F Is the Frobenius norm, v and λ are hyper-parameters, and n is the total number of samples. The data outside the hypersphere in the mapping space is abnormal.
Further, the evaluation value of the data set in the step (3) is the distance between the real-time data and the center of the hypersphere in the mapping space, namely, the anomaly score s (x), and the calculation method is as follows:
s(x)＝‖f(x i ；W * )-c‖ 2
wherein W is * Weights trained for the deep neural network f.
Further, the basis for determining the fault state in step (4) is to determine the data points [ x ] corresponding to the continuous time points at the continuous time points t9 ,x t2 ,…,x tn ]As input to the anomaly detection model, the derived anomaly score [ s ] t) ,s t2 ,…,s tn ]Then calculate the average anomaly score s avr If s avr >R * The device is considered to be in an abnormal state, wherein R * Is the radius of the hypersphere in the mapping space.
The beneficial effects of this application are as follows:
1. the method has the advantages of no need of manually marking data, strong self-adaptive learning capability, strong adaptability, strong robustness and capability of processing large-scale data.
2. In the prior art, a large amount of manual annotation data is needed for training the model, and the method does not need manual annotation data, so that the characteristics and modes can be automatically learned from the data, and a large amount of time and labor cost are saved.
3. Unlike the background artificial marking abnormal data in the prior art, the power plant auxiliary equipment abnormal working condition online detection method based on the unsupervised learning mechanism can adaptively learn and adjust model parameters, so that the method is better suitable for the change of equipment operation environments.
4. In the existing fault detection technology of auxiliary equipment of a power plant, due to the fact that data samples of equipment working states and fault states are very limited, the model is likely to have an overfitting problem, namely the model can only effectively identify known fault samples and cannot be well generalized to unknown fault samples. Meanwhile, the model may have a problem of under fitting, that is, the model cannot effectively identify known fault samples, and cannot generalize to unknown fault samples. The method does not need to know the specific category of the abnormal working condition in advance, can automatically find the abnormal mode and the abnormal working condition from the data, and has better adaptability.
5. The method is not limited by the quality and the number of the labels in the training data, can process the situations of inaccurate, incomplete, missing and the like of the labels, and has stronger robustness.
6. The method can process real-time data of the operation of auxiliary equipment of the power plant on a large scale, and the application range of the method is limited by the scale of the data marked in the prior art.
Drawings
Fig. 1 is a flowchart of a detection method provided by the present invention.
Fig. 2 is a specific embodiment of the detection method provided by the present invention.
Detailed Description
In order to better understand the above technical solution, the following description will be further made with reference to specific embodiments with reference to the accompanying drawings, and it should be noted that the technical solution of the present invention includes, but is not limited to, the following embodiments.
Example 1
As shown in FIG. 1, the method for detecting the abnormal working condition of the auxiliary equipment of the power plant based on the unsupervised learning mechanism comprises the following steps:
step (1): extracting data in a normal state from a database and constructing a training data set; the data in the database in the step (1) comprises real-time operation data and historical operation data of auxiliary equipment of the power plant. The real-time operation data and the historical operation data of the auxiliary equipment of the power plant comprise: (1) basic operational data: power and rotational speed of the device; (2) monitoring data: vibration and sound generated during operation of the device; (3) indirect data: pressure, flow rate, temperature of the fluid in the conduit. For practical auxiliary machine sets, the sensors are limited to be installed for cost and reliability, and only a part of the operation data is usually obtained.
Step (2): extracting features of the training data set and training an anomaly detection model; the feature extracted by the feature extraction in the step (2) includes:
1) Time domain features: mean, variance, minimum, maximum, range, skewness, kurtosis, autocorrelation function, partial autocorrelation function, moving average, moving standard deviation, and exponentially weighted moving average;
2) Frequency domain characteristics: fourier transform coefficients, power spectral density, band energy ratio, frequency entropy;
3) Time-frequency domain characteristics: wavelet transform coefficients, instantaneous frequency, energy spectral density.
And analyzing, weighting and fusing the characteristic values, and forming new and more representative characteristics by fusing a plurality of characteristic values so as to provide more information and further improve the prediction accuracy of the model. Modeling the data of the normal working condition in the historical data obtained from the step (1) by the abnormality detection model in the step (2) based on a One Class Deep SVDD algorithm; the specific method comprises the following steps: training a single class support vector machine model using deep neural network to sample dataFrom the input space-> Is a real set with dimension d and is mapped to the output space of high dimension +.> For a real set of dimensions p, and find a hypersphere in the space, all normal samples are contained, whose objective function is:
wherein x is i For the ith sample, c and R are the center and radius of the hypersphere in the mapping space respectively,for the data expression of the data after the data is mapped by the deep neural network f, L and W are the total layer number and weight of the neural network respectively, W l As the weight of the first layer, II F Is the Frobenius norm, v and λ are hyper-parameters, and n is the total number of samples. The data outside the hypersphere in the mapping space is abnormal.
Step (3): obtaining an evaluation value of the data set by using the anomaly detection model; in the step (3), the evaluation value of the data set is the distance between the real-time data and the center of the hypersphere in the mapping space, namely the abnormal score s (x), and the calculation method is as follows:
s(x)＝‖f(x i ；W * )-c‖ 2
wherein W is * Weights trained for deep neural network fHeavy.
Step (4): and selecting a test data set, taking the test data set as the input of an abnormality detection model, and taking the evaluation values of a plurality of continuous data points as the basis for judging the abnormal working condition of the equipment. The basis for judging the fault state in the step (4) is that the data points [ x ] corresponding to the continuous time points are obtained at the continuous time points t9 ,x t2 ,…,x t< ]As input to the anomaly detection model, the derived anomaly score [ s ] t1 ,s t2 ,…,s tn ]Then calculate the average anomaly score s avr If s avr >R * The device is considered to be in an abnormal state, wherein R * Is the radius of the hypersphere in the mapping space.
Example 2
As shown in FIG. 1, the method for detecting the abnormal working condition of the auxiliary equipment of the power plant based on the unsupervised learning mechanism comprises the following steps:
step (1): establishing a power plant auxiliary equipment operation database including, but not limited to: pressure, flow speed and temperature of the fluid in the pipeline, power and rotating speed of the auxiliary machine unit, and vibration and sound generated in the operation of the auxiliary machine.
Step (2): classifying and summarizing the auxiliary equipment operation data obtained in the step (1): the label D1 represents basic operation parameters of the equipment, namely the power and the rotating speed of the equipment; the label D2 represents fluid parameters in the pipeline, namely the pressure, flow rate and temperature of the fluid when the device is in operation; the label D3 represents the monitored parameters of the device in operation, i.e. vibration, sound generated by the device.
Step (3.1): extracting data from 2 equipment operation databases with labels of D1 and D2 established in the step (2), wherein the data is low-frequency time sequence data, and performing time domain index analysis on the extracted data and extracting characteristic values. The feature values include, but are not limited to: mean, variance, minimum, maximum, range, skewness, kurtosis, autocorrelation function, partial autocorrelation function, moving average, moving standard deviation, and exponentially weighted moving average. And then, respectively adding the extracted characteristic values into a characteristic library labeled as F1 and F2 after stamping the extracted characteristic values with time stamps.
Step (3.2): extracting data from the D3 running database for the label established in the step (2), wherein the data is high-frequency time sequence data, performing time domain index analysis, frequency domain index analysis and time-frequency domain index analysis on the extracted data, and extracting characteristic values. The feature values include, but are not limited to: 1) Time domain features: mean, variance, minimum, maximum, range, skewness, kurtosis, autocorrelation function, partial autocorrelation function, moving average, moving standard deviation, and exponentially weighted moving average; 2) Frequency domain characteristics: fourier transform coefficients, power spectral density, band energy ratio, frequency entropy; 3) Time-frequency domain characteristics: wavelet transform coefficients, instantaneous frequency, energy spectral density. And then adding the extracted characteristic value into a characteristic library with a label of F3 after stamping the time stamp.
Step (4): and (3) analyzing the characteristic values in the characteristic libraries F1, F2 and F3 established in the step (3.1) and the step (3.2), selecting the most relevant or important characteristic, and reducing the characteristic dimension. And meanwhile, weighting and fusing the selected multiple features, and combining the multiple feature values to form new and more representative features so as to provide more information, thereby improving the prediction accuracy of the model, reducing the overfitting risk of the model, improving the robustness of the model and enhancing the interpretability of the features. And (3) selecting, weighting and fusing the characteristic values in the three characteristic libraries of F1, F2 and F3 established in the step (3) to form a new fused characteristic value. And then adding the fusion characteristic value into a characteristic library with a label of F4 after stamping the fusion characteristic value with a time stamp.
Step (5.1): extracting data from the feature library F4 established in the step (4), and establishing a training database; the extraction rule is as follows: and reading the characteristic value and the time stamp, and selecting and extracting data of the time stamp in a normal operation time interval of the equipment.
Step (5.2): based on One Class Deep SVDD algorithm, an anomaly detection model based on an unsupervised learning mechanism is constructed. Specifically, training a single-class support vector machine model by using a deep neural network, taking the characteristic value extracted in the step (4) as an input sample, and taking the input sample from an input spaceAnd mapped to the high-dimensional output space +.>And trying to find a hypersphere in the space, including all normal samples, whose objective function is:
wherein 3 and R are respectively the circle center and the radius of the hypersphere in the mapping space,for data->And the data expression after the mapping of the deep neural network f is that L and W are the total layer number and weight of the neural network respectively, v and lambda are super parameters, and n is the total number of samples. The data outside the hypersphere in the mapping space is abnormal.
Step (6.1): and (3) extracting real-time operation data from the 3 equipment operation databases established in the step (2), and taking the extracted features as input values of an abnormality detection model.
Step (6.2): calculating the distance between the real-time data and the circle center of the hypersphere obtained in the step (5.2) in the mapping space, namely, the anomaly score s (x), wherein the calculation method comprises the following steps:
s(x)＝‖f(x i ；W * )-c‖ 2
wherein W is * Weights trained for the deep neural network f.
Step (6.3): at successive points in time, i.e. [ x t9 ,x t2 ,…,x t< ]Repeating the step (6.1) and the step (6.2) to obtain the corresponding abnormality score s t1 ,s t2 ,…,s tn ]。
Step (6.4): calculating the average value s of the abnormality scores at the successive time points calculated in the step (6.3) avr . If s is avr >R * The equipment unit is considered to be in an abnormal state, wherein R * Is the radius of the super sphere trained in step (5.2) in the mapping space.
As shown in fig. 2, auxiliary equipment operation database 1 in the figure; d1 database stores basic operation parameters 1-1 of the equipment; d2 database stores indirect parameters 1-2 of the device; d3, the database stores monitoring parameters 1-3 in the running process of the equipment; the auxiliary equipment operates the data feature library 2; f1, storing a characteristic value 2-1 of the D1 database in advance according to the method described in the step (2); f2, storing a characteristic value 2-2 of the D2 database in advance according to the method described in the step (2); f3, storing the characteristic value 2-3 of the D3 database in advance according to the method described in the step (2); f4 fusion feature library 3, anomaly detection model training 4, real-time operation feature library 5, and latest operation feature F t0 5-1, next-nearest run feature F t1 5-2, the most recent third bit of operating characteristics F t2 5-3, the operation feature F of the nearest n+1th bit tn 5-4, abnormality detection model 6, abnormality score 7 of auxiliary equipment real-time operation data, latest abnormality score s t0 7-1, next-nearest abnormality score s t1 7-2, anomaly score s of the third most recent bit t2 7-3, abnormality score s at the nearest n+1st position tn 7-4, average anomaly score s avr 8。
Claims (9)
1. The method for detecting the abnormal working condition of the auxiliary equipment of the power plant based on the unsupervised learning mechanism is characterized by comprising the following steps of: the method comprises the following steps:
step (1): extracting data in a normal state from a database and constructing a training data set;
step (2): extracting features of the training data set and training an anomaly detection model;
step (3): obtaining an evaluation value of the data set by using the anomaly detection model;
step (4): and selecting a test data set, taking the test data set as the input of an abnormality detection model, and taking the evaluation values of a plurality of continuous data points as the basis for judging the abnormal working condition of the equipment.
2. The method for detecting abnormal conditions of auxiliary equipment of a power plant based on an unsupervised learning mechanism according to claim 1, wherein the method comprises the following steps: the data in the database in the step (1) comprises real-time operation data and historical operation data of auxiliary equipment of the power plant.
3. The method for detecting abnormal conditions of auxiliary equipment of a power plant based on an unsupervised learning mechanism according to claim 2, wherein the method comprises the following steps: the real-time operation data and the historical operation data of the auxiliary equipment of the power plant comprise: (1) basic operational data: power and rotational speed of the device; (2) monitoring data: vibration and sound generated during operation of the device; (3) indirect data: pressure, flow rate, temperature of the fluid in the conduit.
4. The method for detecting abnormal conditions of auxiliary equipment of a power plant based on an unsupervised learning mechanism according to claim 1, wherein the method comprises the following steps: the feature extracted by the feature extraction in the step (2) includes:
1) Time domain features: mean, variance, minimum, maximum, range, skewness, kurtosis, autocorrelation function, partial autocorrelation function, moving average, moving standard deviation, and exponentially weighted moving average;
2) Frequency domain characteristics: fourier transform coefficients, power spectral density, band energy ratio, frequency entropy;
3) Time-frequency domain characteristics: wavelet transform coefficients, instantaneous frequency, energy spectral density.
5. The method for detecting abnormal conditions of auxiliary equipment of a power plant based on an unsupervised learning mechanism according to claim 4, wherein the method comprises the following steps: and analyzing, weighting and fusing the characteristic values, and forming new and more representative characteristics by fusing a plurality of characteristic values so as to provide more information and further improve the prediction accuracy of the model.
6. The method for detecting abnormal conditions of auxiliary equipment of a power plant based on an unsupervised learning mechanism according to claim 2, wherein the method comprises the following steps: the abnormality detection model in the step (2) models data of normal working conditions in the history data obtained from the step (1) based on a One Class Deep SVDD algorithm.
7. The method for detecting abnormal conditions of auxiliary equipment of a power plant based on an unsupervised learning mechanism according to claim 6, wherein the method comprises the following steps: training a single class support vector machine model using deep neural network to sample dataFrom the input space->Is a real set with dimension d and is mapped to the output space of high dimension +.>For a real set of dimensions p, and find a hypersphere in the space, all normal samples are contained, whose objective function is:
wherein x is i For sample 1, c and R are the center and radius of the hypersphere in the mapping space,for the data expression of the data after the data is mapped by the deep neural network f, L and W are the total layer number and weight of the neural network respectively, W l As the weight of the first layer, II F Is the Frobenius norm, v and λ are hyper-parameters, and n is the total number of samples. The data outside the hypersphere in the mapping space is abnormal.
8. The method for detecting abnormal conditions of auxiliary equipment of a power plant based on an unsupervised learning mechanism according to claim 1, wherein the method comprises the following steps: in the step (3), the evaluation value of the data set is the distance between the real-time data and the center of the hypersphere in the mapping space, namely the abnormal score s (x), and the calculation method is as follows:
s(x)＝‖f(x i ；W * )-c‖ 2
wherein W is * Weights trained for the deep neural network f.
9. The method for detecting abnormal conditions of auxiliary equipment of a power plant based on an unsupervised learning mechanism according to claim 1, wherein the method comprises the following steps: the basis for judging the fault state in the step (4) is that the data points [ x ] corresponding to the continuous time points are obtained at the continuous time points t9 ,x t2 ,…,x t< ]As input to the anomaly detection model, the derived anomaly score [ s ] t1 ,s t2 ,…,s tn ]Then calculate the average anomaly score s avr If s avr >R * The device is considered to be in an abnormal state, wherein R * Is the radius of the hypersphere in the mapping space.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310550153.2A CN116561691A (en) | 2023-05-16 | 2023-05-16 | Power plant auxiliary equipment abnormal condition detection method based on unsupervised learning mechanism |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310550153.2A CN116561691A (en) | 2023-05-16 | 2023-05-16 | Power plant auxiliary equipment abnormal condition detection method based on unsupervised learning mechanism |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116561691A true CN116561691A (en) | 2023-08-08 |
Family
ID=87496130
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310550153.2A Pending CN116561691A (en) | 2023-05-16 | 2023-05-16 | Power plant auxiliary equipment abnormal condition detection method based on unsupervised learning mechanism |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116561691A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118134205A (en) * | 2024-04-30 | 2024-06-04 | 陕西黑石绿能能源科技有限公司 | Hydrogen comprehensive management method and system for hydrogen adding station |
-
2023
- 2023-05-16 CN CN202310550153.2A patent/CN116561691A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118134205A (en) * | 2024-04-30 | 2024-06-04 | 陕西黑石绿能能源科技有限公司 | Hydrogen comprehensive management method and system for hydrogen adding station |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Marugán et al. | Reliability analysis of detecting false alarms that employ neural networks: A real case study on wind turbines | |
| CN111275288B (en) | XGBoost-based multidimensional data anomaly detection method and device | |
| US11840998B2 (en) | Hydraulic turbine cavitation acoustic signal identification method based on big data machine learning | |
| Mathew et al. | Regression kernel for prognostics with support vector machines | |
| CN112836941B (en) | Online health condition assessment method for high-pressure system of steam turbine of coal motor unit | |
| CN111899905A (en) | Fault diagnosis method and system based on nuclear power device | |
| Eickmeyer et al. | Data Driven Modeling for System-Level Condition Monitoring on Wind Power Plants. | |
| CN112990546A (en) | Chemical plant power transformer fault prediction method based on particle swarm and neural network | |
| Lu et al. | Assessment of data-driven, machine learning techniques for machinery prognostics of offshore assets | |
| CN117151684A (en) | Wind power fan data analysis early warning method, system, device and readable storage medium | |
| CN115496108A (en) | Fault monitoring method and system based on manifold learning and big data analysis | |
| CN114004059B (en) | Health portrait method for hydroelectric generating set | |
| CN117930815A (en) | Wind turbine generator remote fault diagnosis method and system based on cloud platform | |
| Bond et al. | A hybrid learning approach to prognostics and health management applied to military ground vehicles using time-series and maintenance event data | |
| 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 | |
| Imamverdiyev et al. | Condition monitoring of equipment in oil wells using deep learning | |
| Wang et al. | Fault detection based on evolving LVQ neural networks | |
| Zhao et al. | Fault Diagnosis of Rolling Bearings based on GA-SVM model | |
| Ayman et al. | Fault Detection in Wind Turbines using Deep Learning | |
| CN118133206B (en) | Sewage treatment fault diagnosis method based on stacking th element learning strategy | |
| CN117892213B (en) | Diagnosis method for icing detection and early warning of wind driven generator blade | |
| Cartelli | A Novel Framework for Condition-Based Maintenance and Performance Analysis Using Data-Driven Approaches | |
| Bondarenko et al. | The Outlook for Machine Learning Application in the Problem of Propulsion Engine Monitoring and Diagnostics | |
| CN117786503A (en) | Cluster balance characterization and space-time collaborative layered monitoring model construction method for wind turbine generator |
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 |