CN114118225A - Method, system, electronic device and storage medium for predicting remaining life of generator - Google Patents
Method, system, electronic device and storage medium for predicting remaining life of generator Download PDFInfo
- Publication number
- CN114118225A CN114118225A CN202111290868.6A CN202111290868A CN114118225A CN 114118225 A CN114118225 A CN 114118225A CN 202111290868 A CN202111290868 A CN 202111290868A CN 114118225 A CN114118225 A CN 114118225A
- Authority
- CN
- China
- Prior art keywords
- data
- generator
- module
- moment
- predicting
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 74
- 238000003860 storage Methods 0.000 title claims abstract description 11
- 238000012549 training Methods 0.000 claims abstract description 26
- 238000007781 pre-processing Methods 0.000 claims abstract description 25
- 238000001914 filtration Methods 0.000 claims description 21
- 238000012544 monitoring process Methods 0.000 claims description 21
- 238000012545 processing Methods 0.000 claims description 19
- 230000008569 process Effects 0.000 claims description 14
- 238000004590 computer program Methods 0.000 claims description 10
- 230000015654 memory Effects 0.000 claims description 9
- 230000004044 response Effects 0.000 claims description 8
- 230000002159 abnormal effect Effects 0.000 claims description 7
- 238000010606 normalization Methods 0.000 claims description 6
- 238000001514 detection method Methods 0.000 abstract description 6
- 239000000284 extract Substances 0.000 abstract description 4
- 238000012423 maintenance Methods 0.000 abstract description 3
- 238000013527 convolutional neural network Methods 0.000 description 50
- 210000004027 cell Anatomy 0.000 description 15
- 238000004422 calculation algorithm Methods 0.000 description 14
- 230000006870 function Effects 0.000 description 11
- 230000007246 mechanism Effects 0.000 description 11
- 230000004913 activation Effects 0.000 description 8
- 238000013528 artificial neural network Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 5
- 238000000605 extraction Methods 0.000 description 5
- 230000000694 effects Effects 0.000 description 4
- 239000011159 matrix material Substances 0.000 description 4
- 238000013459 approach Methods 0.000 description 3
- 230000001413 cellular effect Effects 0.000 description 3
- 238000005259 measurement Methods 0.000 description 3
- 210000002569 neuron Anatomy 0.000 description 3
- 230000002776 aggregation Effects 0.000 description 2
- 238000004220 aggregation Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000004927 fusion Effects 0.000 description 2
- 244000144985 peep Species 0.000 description 2
- 238000011176 pooling Methods 0.000 description 2
- 238000011084 recovery Methods 0.000 description 2
- 230000035945 sensitivity Effects 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 208000000044 Amnesia Diseases 0.000 description 1
- 206010051602 Laziness Diseases 0.000 description 1
- 208000026139 Memory disease Diseases 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 230000003321 amplification Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000005520 cutting process Methods 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 239000006185 dispersion Substances 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 230000006984 memory degeneration Effects 0.000 description 1
- 208000023060 memory loss Diseases 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003199 nucleic acid amplification method Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000036961 partial effect Effects 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 230000036962 time dependent Effects 0.000 description 1
- 239000013598 vector Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Business, Economics & Management (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Quality & Reliability (AREA)
- Probability & Statistics with Applications (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Entrepreneurship & Innovation (AREA)
- Development Economics (AREA)
- Operations Research (AREA)
- Game Theory and Decision Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Primary Health Care (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
The invention relates to a method, a system, an electronic device and a storage medium for predicting the residual life of a generator, comprising the following steps: acquiring historical data of generator equipment; according to historical data, identifying the operation condition of the generator equipment at each moment by adopting a preset condition identification method; preprocessing the operation condition at each moment to obtain a data sample set; training data in the data sample set by adopting an improved CNN module and an improved LSTM module to obtain a hybrid fault prediction model; the improved CNN module is based on attention, and the improved LSTM module is provided with a peephole and a refresh door; and predicting the residual life of the generator based on the mixed fault preset model. The invention overcomes the problems of insensitivity to fault characteristics, multiple parameters and low training speed of the existing prediction method, effectively extracts the multi-parameter characteristics, can accurately predict equipment faults by using multi-characteristic data, reduces the long-time shutdown cost, optimizes the maintenance period and improves early fault detection.
Description
Technical Field
The invention relates to the technical field of equipment fault diagnosis, in particular to a method and a system for predicting the residual life of a generator, electronic equipment and a storage medium.
Background
In terms of the equipment failure prediction method, the prediction model can be classified into a physical model-based, data-driven-based, and hybrid model-based methods. The physical model-based method performs reliability prediction by using a physical model of a machine, requiring a specific model to be built by integrating physical systems and human expertise. The data-driven based approach utilizes condition monitoring data to analyze and predict current and future health conditions, which may be practically applied to non-linear prediction, but requires a relatively higher amount of computation than the physical model-based approach. The hybrid model-based method integrates two methods, namely a physical model and a data driving method, improves the prediction accuracy of the model by integrating the advantages of the two models, but the implementation process is complex, so that the hybrid model-based method is less applied to an actual scene. With the development of technologies such as sensors and data storage, prediction based on data-driven RUL (Remaining Useful Life) has become the current main method. The data-driven methods are mainly neural networks or statistical methods, such as regression models and state space models, which mostly originate from pattern recognition theory. The data-driven approach using artificial intelligence tools does not require in-depth knowledge of the system physics and offers the possibility to use several types of data as input.
The internal structure of the generator is complex, and a large amount of priori knowledge is needed for judging equipment faults in a manual mode, so that a large amount of labeled data cannot be obtained. For large-scale data accumulated by the sensor, the data dimension is large, and label-free data cannot be utilized. Neural Networks have strong nonlinear fitting, feature self-extraction, and feature mapping capabilities, and are well suited to handling such massive state data, and therefore CNNs (Convolutional Neural Networks) are selected as prediction models. However, the convergence rate of the neural network is greatly influenced by the selection of the initial parameters, and the training efficiency of the network can be effectively improved by the appropriate initial parameters. Therefore, there is a need to combine optimization algorithms with CNN in order to get better prediction results.
However, the current hybrid fault prediction model has the following problems:
(1) the partial algorithm still depends on manual feature extraction, an efficient algorithm is needed for automatic feature extraction, the existing prediction model does not consider the time dependence of the sensor monitoring data, and how to utilize the time series characteristics of the data to perform more accurate prediction needs to be considered. In addition, the problem of limited tag data in industrial big data needs to be solved.
(2) The traditional Long Short-Term Memory Network (LSTM) has the defects of excessive parameters, low training speed and the like, and the cell state at the current moment is not determinative to the cell state at the next moment, so that the context correlation of the LSTM on time sequence prediction is cut.
(3) When the traditional CNN is used for prediction, the obtained prediction curve is insensitive to small-range fluctuation of the actual curve, and the prediction algorithm shows inertia of fitting the actual curve.
Disclosure of Invention
The present invention is directed to a method, a system, an electronic device and a storage medium for predicting remaining life of a generator, which are provided to overcome the above-mentioned drawbacks of the prior art.
The technical scheme adopted by the invention for solving the technical problems is as follows: a method for predicting the residual life of a generator is constructed, and comprises the following steps:
acquiring historical data of generator equipment;
according to the historical data, identifying the operation condition of the generator equipment at each moment by adopting a preset condition identification method;
preprocessing the operation condition data at each moment to obtain a data sample set;
training the data in the data sample set by adopting an improved CNN module and an improved LSTM module to obtain a hybrid fault prediction model; the improved CNN module is based on attention, and the improved LSTM module is provided with a peephole and a refresh door;
and predicting the residual life of the generator based on the mixed fault preset model.
In the method for predicting the remaining life of the generator, the identifying the operation condition of the generator equipment at each moment by adopting a preset condition identification method according to the historical data comprises the following steps:
and identifying the operating condition of the generator equipment at each moment by adopting a K-means working condition identification method according to the historical data.
In the method for predicting the remaining life of the generator, the preprocessing the operation condition data at each moment to obtain a data sample set includes:
carrying out abnormal value processing, missing value filling, data normalization processing and signal filtering processing on the operation condition data at each moment;
the signal filtering process includes: and (3) filtering the operation condition data at each moment by adopting a finite impulse response data filter.
In the method for predicting the residual life of the generator, the improved CNN module is a multilayer one-dimensional CNN convolution layer;
the modified LSTM is a stacked multilayer LSTM.
In the method for predicting the remaining life of the generator, the training of the data in the data sample set by using the improved CNN module and the improved LSTM module to obtain the hybrid fault prediction model includes:
extracting generator fault monitoring time series characteristics in the data sample set based on the improved CNN module;
inputting the generator fault monitoring time series signature to the modified LSTM module;
the improved LSTM module is used for extracting and training long-short-term time dependency characteristics of the generator fault monitoring time sequence to obtain the hybrid fault prediction model.
The invention also provides a system for predicting the residual life of the generator, which comprises:
an acquisition unit configured to acquire history data of the generator device;
the identification unit is used for identifying the operation condition of the generator equipment at each moment by adopting a preset condition identification method according to the historical data;
the preprocessing unit is used for preprocessing the operation condition data at each moment to obtain a data sample set;
the model unit is used for training the data in the data sample set by adopting an improved CNN module and an improved LSTM module to obtain a hybrid fault prediction model; the improved CNN module is based on attention, and the improved LSTM module is provided with a peephole and a refresh door;
and the prediction unit is used for predicting the residual life of the generator based on the mixed fault preset model.
In the system for predicting remaining life of a generator according to the present invention, the identification unit includes:
and the working condition identification module is used for identifying the operating working condition of the generator equipment at each moment by adopting a K-means working condition identification method according to the historical data.
In the system for predicting remaining life of a generator according to the present invention, the preprocessing unit includes:
the preprocessing module is used for carrying out abnormal value processing, missing value filling, data normalization processing and signal filtering processing on the operation condition data at each moment;
the signal filtering process includes: and (3) filtering the operation condition data at each moment by adopting a finite impulse response data filter.
The present invention also provides an electronic device comprising: a memory and a processor;
the memory is used for storing a computer program;
the processor is used for executing the computer program to realize the method for predicting the residual life of the generator.
The present invention also provides a storage medium having stored thereon a computer program which, when executed by a processor, causes the processor to process the steps of the above-described method of predicting the remaining life of a generator.
The implementation of the method, the system, the electronic equipment and the storage medium for predicting the residual life of the generator has the following beneficial effects: the method comprises the following steps: acquiring historical data of generator equipment; according to historical data, identifying the operation condition of the generator equipment at each moment by adopting a preset condition identification method; preprocessing the operation condition at each moment to obtain a data sample set; training data in the data sample set by adopting an improved CNN module and an improved LSTM module to obtain a hybrid fault prediction model; the improved CNN module is based on attention, and the improved LSTM module is provided with a peephole and a refresh door; and predicting the residual life of the generator based on the mixed fault preset model. The invention overcomes the problems of insensitivity to fault characteristics, multiple parameters and low training speed of the existing prediction method, thereby effectively extracting the multi-parameter characteristics, accurately predicting the equipment fault by utilizing multi-characteristic data, reducing the long-time shutdown cost, optimizing the maintenance period and improving the early fault detection.
Drawings
The invention will be further described with reference to the accompanying drawings and examples, in which:
FIG. 1 is a schematic flow chart of a method for predicting the remaining life of a generator according to an embodiment of the present invention;
fig. 2 is a schematic structural diagram of a CNN module based on attention according to an embodiment of the present invention;
FIG. 3 is a schematic structural diagram of an LSTM module with peep hole and refresh door according to an embodiment of the present invention;
fig. 4 is a schematic block diagram of a system for predicting the remaining life of a generator according to an embodiment of the present invention.
Detailed Description
For a more clear understanding of the technical features, objects and effects of the present invention, embodiments of the present invention will now be described in detail with reference to the accompanying drawings.
The method for predicting the residual life of the generator provided by the invention comprises the steps of firstly improving the traditional Convolutional Neural Networks (CNN), providing the CNN based on an attention mechanism, extracting the characteristics of a generator fault monitoring time sequence, then removing an input gate of a long-short term memory Network (LSTM), replacing a forgetting gate with an updating gate, obtaining an LSTM module with a peephole and the updating gate, generating a characteristic description by using the LSTM module with the peephole and the updating gate, combining the improved CNN module with the improved LSTM module to form a mixed fault prediction module, and finally predicting the service life of the generator by using the mixed fault prediction module. According to the invention, through a prediction algorithm of a mixed CNN model and an LSTM model, attention is focused on fault characteristics, and the problems of 'laziness' (insensitivity to fault characteristics), more parameters, low training speed and the like in the conventional prediction are solved, so that multi-parameter characteristics are effectively extracted. The multi-feature data can be used for accurately predicting equipment faults, reducing long-time shutdown cost, optimizing maintenance period and improving early fault detection.
Specifically, as shown in fig. 1, the method for predicting the remaining life of the generator includes the following steps:
and step S101, acquiring historical data of the generator equipment.
In the embodiment of the invention, the historical data of the generator equipment includes, but is not limited to, state monitoring data such as pressure data, vibration data, rotating speed data, temperature data and the like of the generator equipment. Wherein these condition monitoring data may be obtained via corresponding sensors and/or database storage.
And S102, identifying the operation condition of the generator equipment at each moment by adopting a preset condition identification method according to the historical data.
It can be understood that the complexity and the intelligent program of the generator equipment structure are continuously improved, and the corresponding equipment working mode or the operating condition is more complicated and changeable. In the embodiment of the invention, the adverse effect caused by complex working conditions can be eliminated in the data preprocessing stage by firstly identifying the working conditions of the equipment and then preprocessing the data according to the identification result.
Optionally, in the embodiment of the present invention, identifying the operating condition of the generator device at each moment by using a preset condition identification method according to the historical data includes: and identifying the operating condition of the generator equipment at each moment by adopting a K-means working condition identification method according to historical data. Specifically, the working condition recognition method of the K-means is adopted for recognizing the working condition of the equipment, and the operation working condition of the equipment at each moment in the multidimensional sensor sequence can be recognized after unsupervised clustering training.
In the embodiment of the invention, in the process of identifying the K-means working condition, the input historical data is an original multi-dimensional sensing sequence which is a time sequence, and the K-means algorithm is utilized to identify the working condition aiming at the data of the equipment at each moment.
In the embodiment of the invention, the distance measurement in the K-means working condition identification algorithm adopts the Euclidean distance, and the clustering center adopts random sampling initialization. The method comprises the following specific steps:
a1, randomly selecting K samples as initial clustering centers, as shown in formula (1):
wherein (1) Z in the formulaj kIndicating that k initial cluster centers Xj indicate the jth sample.
a2, classifying the categories according to the distance from each sample to the cluster center, as shown in formula (2):
wherein in the formula (2), d (X1, Z)j k) Indicating the distance of the ith sample from the jth cluster center.
a3, recalculating the cluster centers, as shown in formula (3):
wherein in the formula (3), Zj k+1After the average value is calculated after the clustering is completed, the average value is used as a new clustering center.
a4, determining whether the stop condition is reached, i.e. the cluster center is hardly changed, as shown in equation (9). And directly returning the result if the stopping condition is reached, otherwise, continuing the steps a2 and a 3.
and a5, obtaining a clustering result to obtain a clustering center.
Further, defining the generator device at tiThe sensor monitoring data at the moment is xiThen { x } represents the entire sensor sequence, the set of K-class operating conditions is Ω ═ 1, 2i) Identifying and recording the operation condition of the equipment at each momentTo representDevice tiThe time belongs to the working condition k. And respectively carrying out z-score standardization on each operating condition, wherein the corresponding mean value and standard deviation are respectively shown as a formula (4) and a formula (5).
s(k)=Std({x}(k)) (5)。
And S103, preprocessing the operation condition data at each moment to obtain a data sample set.
Optionally, in the embodiment of the present invention, the preprocessing the operation condition data at each time to obtain the data sample set includes: carrying out abnormal value processing, missing value filling, data normalization processing and signal filtering processing on the operation condition data at each moment; the signal filtering process includes: and (3) filtering the operation condition data at each moment by adopting a finite impulse response data filter.
In the embodiment of the invention, the original data of the operation condition at each moment is cleaned by preprocessing the operation condition data at each moment, so that the interference of abnormal data is reduced.
Specifically, in step S103, any time x in the raw sensor monitoring dataiAfter the working condition identification and data preprocessing processes, converting the working condition identification and data preprocessing processes into yiThe specific process is as follows:
(6) in the formula, yiRepresents tiRecognition of the operating conditions at the moment, xni、And Sn (k)Are each xi、And s(k)The nth element of (1).
Furthermore, in the embodiment of the present invention, a finite impulse response digital filter (finite impulse response digital filter) is adopted, because the digital filter has a full zero structure, the stability problem does not exist, the digital filter has the advantages of ensuring any amplitude-frequency characteristic, having strict linear phase-frequency characteristic, being capable of utilizing fast Fourier transform to accelerate filtering operation, and the like, and occupies a very important position in digital filtering. The fir filter has the following form:
(7) where N is the filter length, x (k) is the value of the sensor measurement signal at time k, y (k) is the filtered signal value at time k, { b (i) }, i ═ 1, 2.. and N } represent a series of weighting coefficients that determine the filter characteristics, and the sum is 1. Wherein, when the values of { b (i), i ═ 1, 2.
And step S104, training the data in the data sample set by adopting an improved CNN module and an improved LSTM module to obtain a hybrid fault prediction model.
In some embodiments, training data in the data sample set using the modified CNN module and the modified LSTM module to obtain the hybrid fault prediction model includes: extracting generator fault monitoring time sequence characteristics in a data sample set based on an improved CNN module; inputting generator fault monitoring time series characteristics to an improved LSTM module; the improved LSTM module is used for carrying out long-short-term time dependency feature extraction and training on the generator fault monitoring time sequence to obtain a hybrid fault prediction model.
Optionally, in the embodiment of the present invention, the improved CNN module is a CNN module based on attention. Specifically, the improved CNN module of the embodiment of the present invention is a multilayer one-dimensional CNN convolutional layer. Wherein, the multilayer one-dimensional CNN convolution layer carries out convolution calculation along the time axis direction, the time window size is within the range of 15-30, and the convolution kernel size is set to be 3-7. Wherein, stackingThe multi-layer one-dimensional CNN convolutional layer of (a) uses zero padding to keep the time window size unchanged, wherein the output format of the improved CNN module is: n is a radical ofwindow×NID-CNNAnd can be directly input into the improved LSTM module.
Specifically, as shown in fig. 2, a schematic structural diagram of a CNN module based on an attention mechanism according to an embodiment of the present invention is shown.
As shown in fig. 2, assuming that m time series of length n are input, the size can be expressed as (n × m). The CNN module of the attention mechanism is composed of two parts of feature aggregation and scale recovery. The feature aggregation part extracts key features from the cross-scale subsequence by using the stacking of multiple layers of convolution and pooling layers, and the last layer excavates a linear relation by using a convolution kernel with the size of 1 multiplied by 1. And the scale recovery part recovers the key features to (n multiplied by m), namely the key features are consistent with the size of the output features of the CNN module, and the value is constrained between 0 and 1 by applying a Sigmoid function. And taking the extracted context features as a significance feature graph of the basic features. The input of each parallel module has adjustable overlap, the model can better adapt to data with different laws by adjusting the input overlap step length, the length of one-time integral input of the combined model can be further enlarged, and more accurate long-term characteristics can be captured. The input between the parallel modules is not densely overlapped, so that overfitting can be avoided.
Based on the CNN module of the attention mechanism, the branch input sequence of the attention module covers the input of the CNN and stacks the deep convolutional network and the pooling layer, the input feeling corresponding to the features is also enlarged, and the model obtains more comprehensive context information, so that the importance degree of the features of the current local sequence is learned. Through the attention module, the influence degree of important time sequence characteristics on the model can be improved, the interference of non-important characteristics on the model is inhibited, and the problem that the model cannot distinguish the difference of the importance degrees of the time sequence data is effectively solved. Meanwhile, the standard CNN branch and the attention branch take sequences with different lengths as input, and multi-scale input can effectively extract richer short sequence features and can prevent the problems that input cutting is too small or too large in single-scale CNN and matching features are lost in prediction.
The higher the importance of the output characteristics of the CNN module is, the closer the output of the CNN module corresponding to the attention mechanism module is to 1; conversely, the lower the importance of the CNN module output feature, the closer its output to 0 corresponding to the attention mechanism module. The degree of importance of the features is reflected by the height of the numerical value, so that the identification of the important features is completed. The fused result is used as the input of one node of the LSTM module unit, and the LSTM is used for sequence modeling to obtain a final prediction result. And an LSTM module is used for extracting coarse-grained features from the remarkable fine-grained features extracted from the CNN based on the attention mechanism, and the problems of memory loss and gradient dispersion caused by overlong step length are prevented while the dimensional features are finely processed. The system structure can capture the time dependency of effective features extracted after the convolution operation is optimized by an attention mechanism, realizes the fusion of coarse and fine granularity features and comprehensively describes time sequence data.
In the embodiment of the invention, the improved LSTM module is an LSTM module with a peephole and a refresh door. Wherein the modified LSTM is a stacked multilayer LSTM.
As shown in fig. 3, an LSTM module with peep hole and refresh door is provided for an embodiment of the present invention. Wherein, Ct-1The cellular state is the state of the cell at the previous moment, the state of the cell is the state of the memory of the neuron to the past moment, and the LSTM neural network completes the memory and transmission of the time sequence characteristics through the cell state. h ist-1Is the output of the previous time, xtIs an input of the current time, CtIs the cell state at the present time, htIs the output of the current time.
In the embodiment of the invention, the improved LSTM removes an input door, replaces a forgetting door with an updating door, and updates the door utFrom the current time input xtOutput h at the last momentt-1And the cellular state C of the last momentt-1Determining that the updated gate weight is multiplied by an amplification matrix formed by the three parameters and added with a bias matrix, outputting a number between 0 and 1 through a Sigmoid activation function, wherein '0' represents that the cell state at the previous moment is completely abandoned, and '1' represents that the cell state at the previous moment is completely reserved, and the formula of the updated gate is as follows:
ut=σ(Wu·[Ct-1,ht-1,xt]+bu) (8)。
wherein, in the formula (8), sigma is sigmoid activation function, buIs the bias.
The function of the "input gate" of the traditional LSTM is now replaced by an update gate, which determines how much of the current time input the cell state receives, with a value of 1-ut. Input state at present timeThe following were used:
wherein in the formula (9), WiIs a weight matrix of the input gate, biIs the corresponding offset of the input gate.
The cell state at the current time is determined by the update gate, and the formula of the cell state at the current time is as follows:
the output gate is similar to the input gate, and x is input from the current timetOutput h at the last momentt-1And the cellular state C of the last momentt-1Together, the output gate controls how much of the cell state at the current time flows into the next time as an output, and the output gate formula is as follows:
ot=σ(W0·[ht-1,xt]+b0) (11)。
wherein in the formula (11), W0Is the weight matrix of the output gate, σ is the sigmoid activation function, b0Is the corresponding input bias.
Finally, the cell state at the current time passes through the tanh activation function and is output at the output gate otUnder the control of (3), output h to the next momenttThe output at the current time is as follows:
ht=tanh(Ct)*ot (12)。
the embodiment of the invention adopts the overlapped multilayer LSTM module and is mainly responsible for extracting the time-dependent characteristics. To prevent over-fitting the training sample set, Dropout regularization technique is applied to each LSTM layer, i.e. there is a certain probability in the training to ignore part of the neurons, and the parameter size is set to 0.2-0.5. The LSTM layer then concatenates the feature vectors to the fully-connected layer for regression prediction of RUL values. All the neural network layers adopt a ReLU function as an activation function, the Re LU is a newer activation function, gradient descent and error back propagation can be effectively carried out through the activation function, the problem of gradient disappearance is avoided, in addition, the generalization speed of the Re LU is higher, and the convergence speed of the gradient descent can be accelerated. The formula for the activation function is as follows:
fReLU(x)=Max(0,x) (13)。
the model loss is calculated using equation (12) and network parameters are optimized using the Adam algorithm based on the error.
Ltotal(θ)=αLscore(θ)+(1-α)LMSE(θ)+J(θ) (14)。
Wherein theta represents a network weight parameter, J (theta) represents a regularization term, and alpha represents Lscore(theta) as a hyperparameter of the model.
And S105, predicting the residual service life of the generator based on the mixed fault preset model.
The method for predicting the residual life of the generator comprises two stages of online application and offline modeling, wherein an online part and an offline part are mutually supplemented and mutually supported. In the off-line stage, test data, historical monitoring data and the like of the equipment system are fully utilized, and a residual life prediction model of the generator is established; in the online stage, after necessary data preprocessing and other operations are carried out on the data collected by real-time monitoring, the remaining life of the generator is predicted by means of an intelligent algorithm according to the actual operating environment (such as external environment, load condition and the like) of the generator equipment. In the framework, the off-line stage training learning and the on-line stage actual prediction are fully combined, the model can be adjusted in an off-line mode by referring to the prediction effect in the actual on-line application, and then a closed-loop system from off-line stage modeling, training and learning to on-line stage actual application is constructed, and then the algorithm model is dynamically optimized according to the on-line actual application effect.
Further, in the embodiment of the present invention, a time window sliding structure is adopted to construct a training sample during training, the step length is 1, and the size of the time window is recorded as NwindowSensor data dimension is noted as NsensorIf the input format is Nwindow×NsensorAnd after the original monitoring data are sliced along a time axis, the original monitoring data are directly input into the mixed model, and the characteristics do not need to be manually extracted by depending on the technologies such as field knowledge or signal processing and the like.
In one embodiment, the improved CNN module may be designed with three layers, each layer having 32, 48, 64 convolution kernels and a size of 2. Wherein, more time characteristics can be extracted by changing the total area core number in the improved CNN module. In addition, the depth of the network can be increased by increasing the number of LSTM network layers, and the prediction capability of the prediction model is improved. Alternatively, the present invention may create two layers of LSTM, each layer having 32, 48 neuron numbers.
The prediction method provided by the embodiment of the invention utilizes the advantages of the strong characteristic self-extraction capability of the CNN and the LSTM time sequence sensitivity, can fully extract the characterization capability of data in two dimensions of space and time, better fits the nonlinear relation between the residual service life and the monitoring variable in a short period, and the prediction result is closer to the actual measurement result, thereby being a more advanced model for predicting the residual service life of the generator.
Further, on the basis of the traditional LSTM, a door updating mechanism with a peephole is provided. The updating door combining the forgetting door and the output door can effectively reduce the parameter scale and improve the training speed. The peephole mechanism can enable the cell state at the current moment to participate in formulating the cell state at the next moment, further strengthens the relation of time sequence data context, and improves the accuracy of prediction.
In addition, CNN based attention mechanisms can increase the sensitivity of the algorithm to fault signatures, focusing attention on local sequences. In addition, the improved CNN can perform necessary filtering on most normal data, so that the algorithm focuses more on the fluctuation rule of the local sequence, and the features of the local sequence are effectively extracted.
Meanwhile, in the problem of time series data anomaly detection, an anomaly detection algorithm adopting fusion of two models is a large trend of time series data anomaly detection. The hybrid fault prediction model applies the deep learning idea to industrial big data, fully utilizes the label-free data, and improves the prediction capability of a fault prediction algorithm by learning the faults of the generator equipment.
Fig. 4 is a schematic block diagram of a system for predicting the remaining life of a generator according to the present invention. The system for predicting the residual life of the generator can be used for realizing the method for predicting the residual life of the generator disclosed by the embodiment of the invention.
As shown in fig. 4, the system for predicting the remaining life of the generator includes:
an obtaining unit 401 is configured to obtain history data of the generator device.
The identifying unit 402 is configured to identify an operating condition of the generator device at each moment by using a preset condition identifying method according to the historical data.
The preprocessing unit 403 is configured to preprocess the operation condition data at each time to obtain a data sample set.
And the model unit 404 is configured to train data in the data sample set by using the improved CNN module and the improved LSTM module to obtain a hybrid fault prediction model. The improved CNN module is an attention-based CNN module, and the improved LSTM module is an LSTM module with a peephole and a refresh door.
And the prediction unit 405 is used for predicting the residual life of the generator based on the hybrid fault preset model.
Specifically, in some embodiments, the identifying unit 402 includes: and the working condition identification module is used for identifying the operating working condition of the generator equipment at each moment by adopting a K-means working condition identification method according to historical data.
In some embodiments, the pre-processing unit 403 includes: and the preprocessing module is used for carrying out abnormal value processing, missing value filling, data normalization processing and signal filtering processing on the operation condition data at each moment. Wherein the signal filtering process includes: and (3) filtering the operation condition data at each moment by adopting a finite impulse response data filter.
The present invention also provides an electronic device comprising: a memory and a processor.
The memory is for storing a computer program.
The processor is used for executing the computer program to realize the method for predicting the residual life of the generator disclosed by the embodiment of the invention.
The invention also provides a storage medium, wherein a computer program is stored on the storage medium, and when the computer program is executed by the processor, the processor is enabled to process the steps of the method for predicting the residual life of the generator disclosed by the embodiment of the invention.
The above embodiments are merely illustrative of the technical ideas and features of the present invention, and are intended to enable those skilled in the art to understand the contents of the present invention and implement the present invention, and not to limit the scope of the present invention. All equivalent changes and modifications made within the scope of the claims of the present invention should be covered by the claims of the present invention.
Claims (10)
1. A method for predicting the residual life of a generator is characterized by comprising the following steps:
acquiring historical data of generator equipment;
according to the historical data, identifying the operation condition of the generator equipment at each moment by adopting a preset condition identification method;
preprocessing the operation condition data at each moment to obtain a data sample set;
training the data in the data sample set by adopting an improved CNN module and an improved LSTM module to obtain a hybrid fault prediction model; the improved CNN module is based on attention, and the improved LSTM module is provided with a peephole and a refresh door;
and predicting the residual life of the generator based on the mixed fault preset model.
2. The method for predicting the remaining life of the generator according to claim 1, wherein the step of identifying the operation condition of the generator equipment at each moment by adopting a preset condition identification method according to the historical data comprises the following steps:
and identifying the operating condition of the generator equipment at each moment by adopting a K-means working condition identification method according to the historical data.
3. The method for predicting the remaining life of the generator according to claim 1, wherein the preprocessing the operation condition data at each moment to obtain a data sample set comprises:
carrying out abnormal value processing, missing value filling, data normalization processing and signal filtering processing on the operation condition data at each moment;
the signal filtering process includes: and (3) filtering the operation condition data at each moment by adopting a finite impulse response data filter.
4. The method for predicting remaining life of power generator as claimed in claim 1, wherein said modified CNN module is a multi-layer one-dimensional CNN convolutional layer;
the modified LSTM is a stacked multilayer LSTM.
5. The method for predicting the remaining life of the generator according to claim 1, wherein the training of the data in the data sample set by using the modified CNN module and the modified LSTM module to obtain the hybrid fault prediction model comprises:
extracting generator fault monitoring time series characteristics in the data sample set based on the improved CNN module;
inputting the generator fault monitoring time series signature to the modified LSTM module;
the improved LSTM module is used for extracting and training long-short-term time dependency characteristics of the generator fault monitoring time sequence to obtain the hybrid fault prediction model.
6. A system for predicting remaining life of a generator, comprising:
an acquisition unit configured to acquire history data of the generator device;
the identification unit is used for identifying the operation condition of the generator equipment at each moment by adopting a preset condition identification method according to the historical data;
the preprocessing unit is used for preprocessing the operation condition data at each moment to obtain a data sample set;
the model unit is used for training the data in the data sample set by adopting an improved CNN module and an improved LSTM module to obtain a hybrid fault prediction model; the improved CNN module is based on attention, and the improved LSTM module is provided with a peephole and a refresh door;
and the prediction unit is used for predicting the residual life of the generator based on the mixed fault preset model.
7. The generator remaining life prediction system of claim 6, wherein the identification unit comprises:
and the working condition identification module is used for identifying the operating working condition of the generator equipment at each moment by adopting a K-means working condition identification method according to the historical data.
8. The generator remaining life prediction system of claim 6, wherein the preprocessing unit comprises:
the preprocessing module is used for carrying out abnormal value processing, missing value filling, data normalization processing and signal filtering processing on the operation condition data at each moment;
the signal filtering process includes: and (3) filtering the operation condition data at each moment by adopting a finite impulse response data filter.
9. An electronic device, comprising: a memory and a processor;
the memory is used for storing a computer program;
the processor is configured to execute the computer program to implement the method of predicting remaining life of a generator according to any one of claims 1 to 5.
10. A storage medium having stored thereon a computer program which, when executed by a processor, causes the processor to process the steps of the method of predicting remaining life of a generator according to any one of claims 1 to 5.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111290868.6A CN114118225A (en) | 2021-11-02 | 2021-11-02 | Method, system, electronic device and storage medium for predicting remaining life of generator |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111290868.6A CN114118225A (en) | 2021-11-02 | 2021-11-02 | Method, system, electronic device and storage medium for predicting remaining life of generator |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114118225A true CN114118225A (en) | 2022-03-01 |
Family
ID=80380244
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111290868.6A Pending CN114118225A (en) | 2021-11-02 | 2021-11-02 | Method, system, electronic device and storage medium for predicting remaining life of generator |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114118225A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114980723A (en) * | 2022-05-24 | 2022-08-30 | 中国科学技术大学 | Fault prediction method and system for cross-working-condition chip mounter suction nozzle |
CN115482665A (en) * | 2022-09-13 | 2022-12-16 | 重庆邮电大学 | Knowledge and data collaborative driving multi-granularity traffic accident prediction method and device |
CN115759357A (en) * | 2022-10-29 | 2023-03-07 | 思维实创(哈尔滨)科技有限公司 | Power supply equipment safety prediction method, system, equipment and medium based on PSCADA data |
CN117368724A (en) * | 2023-12-08 | 2024-01-09 | 天津国能津能滨海热电有限公司 | Motor life prediction method, device, medium and equipment |
-
2021
- 2021-11-02 CN CN202111290868.6A patent/CN114118225A/en active Pending
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114980723A (en) * | 2022-05-24 | 2022-08-30 | 中国科学技术大学 | Fault prediction method and system for cross-working-condition chip mounter suction nozzle |
CN115482665A (en) * | 2022-09-13 | 2022-12-16 | 重庆邮电大学 | Knowledge and data collaborative driving multi-granularity traffic accident prediction method and device |
CN115482665B (en) * | 2022-09-13 | 2023-09-15 | 重庆邮电大学 | Knowledge and data collaborative driving multi-granularity traffic accident prediction method and device |
CN115759357A (en) * | 2022-10-29 | 2023-03-07 | 思维实创(哈尔滨)科技有限公司 | Power supply equipment safety prediction method, system, equipment and medium based on PSCADA data |
CN117368724A (en) * | 2023-12-08 | 2024-01-09 | 天津国能津能滨海热电有限公司 | Motor life prediction method, device, medium and equipment |
CN117368724B (en) * | 2023-12-08 | 2024-03-19 | 天津国能津能滨海热电有限公司 | Motor life prediction method, device, medium and equipment |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN114118225A (en) | Method, system, electronic device and storage medium for predicting remaining life of generator | |
CN116757534B (en) | Intelligent refrigerator reliability analysis method based on neural training network | |
US10996664B2 (en) | Predictive classification of future operations | |
CN113486578B (en) | Method for predicting residual life of equipment in industrial process | |
CN113255848A (en) | Water turbine cavitation sound signal identification method based on big data learning | |
CN118154174B (en) | Intelligent operation and maintenance cloud platform for industrial equipment | |
CN114218872B (en) | DBN-LSTM semi-supervised joint model-based residual service life prediction method | |
CN116929815A (en) | Equipment working state monitoring system and method based on Internet of things | |
CN114282443A (en) | Residual service life prediction method based on MLP-LSTM supervised joint model | |
CN115659174A (en) | Multi-sensor fault diagnosis method, medium and equipment based on graph regularization CNN-BilSTM | |
CN112504682A (en) | Chassis engine fault diagnosis method and system based on particle swarm optimization algorithm | |
CN115496384A (en) | Monitoring management method and device for industrial equipment and computer equipment | |
CN116738868A (en) | Rolling bearing residual life prediction method | |
CN117032165A (en) | Industrial equipment fault diagnosis method | |
CN113935413A (en) | Distribution network wave recording file waveform identification method based on convolutional neural network | |
CN117154256A (en) | Electrochemical repair method for lithium battery | |
CN117369418A (en) | Fault diagnosis method, system, storage medium and equipment for feeding system of numerical control machine tool | |
CN114764538A (en) | Equipment sound signal pattern recognition model | |
CN116861232A (en) | Air quality data anomaly detection model based on DBN-OCSVM | |
CN115794805A (en) | Medium-low voltage distribution network measurement data supplementing method | |
CN114298413A (en) | Hydroelectric generating set runout trend prediction method | |
CN115081514A (en) | Industrial equipment fault identification method under data imbalance condition | |
Ge et al. | Remaining useful life prediction using deep multi-scale convolution neural networks | |
CN117370870B (en) | Knowledge and data compound driven equipment multi-working condition identification and performance prediction method | |
CN115556099B (en) | Sustainable learning industrial robot fault diagnosis system and method |
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 |