CN114611803A - Switch device service life prediction method based on degradation characteristics - Google Patents
Switch device service life prediction method based on degradation characteristics Download PDFInfo
- Publication number
- CN114611803A CN114611803A CN202210255617.2A CN202210255617A CN114611803A CN 114611803 A CN114611803 A CN 114611803A CN 202210255617 A CN202210255617 A CN 202210255617A CN 114611803 A CN114611803 A CN 114611803A
- Authority
- CN
- China
- Prior art keywords
- lstm model
- degradation
- switching device
- loss value
- prediction method
- 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
Images
Classifications
-
- 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
- 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
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biophysics (AREA)
- Strategic Management (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Game Theory and Decision Science (AREA)
- Development Economics (AREA)
- Entrepreneurship & Innovation (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention discloses a switch device service life prediction method based on degradation characteristics, which comprises the following steps: s1, collecting degradation characteristic parameter information of a switch device; s2, constructing an initialized LSTM model; s3, making the degeneration characteristic parameter information into training sample data conforming to an input format of an LSTM model; s4, inputting training sample data into the LSTM model, adjusting target parameter values in the LSTM model to enable the residual life loss value output by the LSTM model to be minimum, and taking the LSTM model set when the loss value is minimum as the well-trained LSTM model; and S5, inputting the degradation characteristic parameter information of the switching device to be tested into the well-trained LSTM model, and outputting the predicted value of the residual life of the switching device to be tested. The invention can reflect the device degradation rule, avoids the complication of model parameters and has low dependence on data quantity.
Description
Technical Field
The invention relates to the field of switching devices, in particular to a switching device service life prediction method based on degradation characteristics.
Background
At present, two main methods, namely a model driving method and a data driving method, are mainly adopted for predicting the service life of a switching device.
The model driving method cannot reflect the degradation characteristic rule, usually needs to rely on accelerated degradation test and finite element simulation to obtain model parameters, the parameter obtaining process needs to spend high economy and time cost, the model can only be used under specific working conditions, and the application range is small.
The data driving method is divided into two types, wherein the prediction precision of the statistical sequence method is related to historical data, a probability density function and parameter selection, and the statistical sequence method has larger uncertainty margin; the machine learning method needs a large amount of raw data to mine the relationship between the historical data and the service life, and the raw data is high in economic and time cost in actual situations, so that the actual application requirements are difficult to meet.
Therefore, in order to solve the above problems, a method for predicting the lifetime of a switching device, which has a wide application range, a low data amount dependency, and high efficiency, is required.
Disclosure of Invention
In view of this, the present invention aims to overcome the defects in the prior art, and provide a method for predicting the lifetime of a switching device based on degradation characteristics, which can reflect the degradation rule of the device, avoid the complication of model parameters, and have low dependence on data amount.
The invention discloses a method for predicting the service life of a switching device based on degradation characteristics, which comprises the following steps:
s1, collecting degradation characteristic parameter information of a switch device;
s2, constructing an initialized LSTM model;
s3, making the degeneration characteristic parameter information into training sample data conforming to an input format of an LSTM model;
s4, inputting training sample data into the LSTM model, adjusting target parameter values in the LSTM model to enable the residual life loss value output by the LSTM model to be minimum, and taking the LSTM model set when the loss value is minimum as the well-trained LSTM model; the target parameters comprise the number of hidden layer layers, the number of hidden layer units and an initial learning rate;
and S5, inputting the degradation characteristic parameter information of the switching device to be tested into the well-trained LSTM model, and outputting the predicted value of the residual life of the switching device to be tested.
Further, the step S1 further includes: and carrying out noise reduction processing on the degradation characteristic parameter information to obtain the degradation characteristic parameter information after noise reduction.
Further, adjusting the target parameter value in the LSTM model to minimize the remaining life loss value output by the LSTM model specifically includes:
s41, setting the maximum optimization times N, randomly initializing target parameters, taking the initialized target parameters as initial sampling points, and inputting the initial sampling points into Gaussian regression;
s42, correcting the Gaussian regression according to the residual life loss value output by the LSTM model, so that the Gaussian regression is close to the distribution of a real function;
s43, selecting a next sampling point to be evaluated from the corrected Gaussian regression by using an acquisition function, transmitting the sampling point as an input into an LSTM model for training, and correcting the Gaussian regression again;
s44, judging whether the correction times reach the optimization times N, if not, repeating the steps S42-S43; and if so, outputting a sampling point which enables the residual life loss value to be minimum.
Further, step S41 includes: setting a maximum optimization time T;
in step S44, the method further includes: judging whether the execution time of the steps S42-S43 is greater than the optimization time T, if not, repeating the steps S42-S43; and if so, outputting a sampling point which enables the residual life loss value to be minimum.
Further, the probability distribution of the optimization process is determined according to the following formula:
wherein p (E | D) is the posterior probability,is the probability coefficient, p (E) is the prior probability; e is the distribution of sampling points and D is the evaluation of the acquisition function.
Further, a sliding window method is adopted to make training sample data.
Further, the loss value is calculated using a root mean square difference.
The invention has the beneficial effects that: the invention discloses a switch device life prediction method based on degradation characteristics, which comprises the steps of constructing an LSTM model, inputting degradation characteristic parameters of a switch device into the LSTM model as training samples, optimizing the LSTM model by Bayesian optimization to obtain an optimized LSTM model, and finally predicting the residual life of a switch device to be tested by using the optimized LSTM model. The method can reflect the device degradation rule, avoids the complication of model parameters, has low dependence degree on data quantity, and is suitable for scenes in which a physical model of a prediction object cannot be built or is too complex and difficult to build.
Drawings
The invention is further described below with reference to the following figures and examples:
FIG. 1 is a flow chart of a prediction method according to the present invention;
FIG. 2 is a diagram of the cell structure of the LSTM neural network of the present invention;
FIG. 3 is a schematic diagram of a sliding window method for constructing a sample sequence according to the present invention;
FIG. 4 is a visual schematic diagram of a Bayesian optimization process of the present invention;
FIG. 5 is a diagram illustrating performance indicators of multiple prediction models under average absolute error according to the present invention;
FIG. 6 is a diagram illustrating performance metrics of multiple prediction models under root mean square error according to the present invention;
FIG. 7 is a diagram of the performance index of the prediction models in goodness of fit.
Detailed Description
The invention is further described with reference to the accompanying drawings, in which:
the invention discloses a method for predicting the service life of a switching device based on degradation characteristics, which comprises the following steps:
s1, collecting degradation characteristic parameter information of a switch device; the degradation characteristic parameters comprise voltage, current and temperature, and any one of the degradation characteristic parameters can be selected as data for collection, for example, the voltage of a switching device is collected as test data;
s2, constructing an initialized LSTM model; the LSTM is used as a special Recurrent Neural Network (RNN), solves the problems of gradient disappearance and gradient explosion of the traditional RNN, has better long-term prediction capability, and is very suitable for the service life prediction of a switching device;
s3, making the degeneration characteristic parameter information into training sample data conforming to an input format of an LSTM model;
s4, inputting training sample data into the LSTM model, adjusting target parameter values in the LSTM model to enable the residual life loss value output by the LSTM model to be minimum, and taking the LSTM model set when the loss value is minimum as the well-trained LSTM model; the target parameters comprise the number of hidden layer layers, the number of hidden layer units and an initial learning rate; the residual life loss value represents the difference between the predicted residual life and the real residual life output by the LSTM model, the real residual life information can be obtained by calculating actual degradation characteristic parameters, such as the voltage characteristic of the switching device, the real residual life of the switching device is obtained by counting the life stage of the current voltage and combining the life cycle; the loss value may be calculated using a root mean square difference;
and S5, inputting the degradation characteristic parameter information of the switching device to be tested into the well-trained LSTM model, and outputting the predicted value of the residual life of the switching device to be tested.
In this embodiment, the step S1 further includes: and carrying out noise reduction processing on the degradation characteristic parameter information to obtain the degradation characteristic parameter information after noise reduction. Specifically, the Gaussian smoothing algorithm is used for reducing the noise of the degradation characteristic parameter information, and the LSTM network is prevented from being sensitive to the noise of input data, so that the prediction precision is improved. The gaussian smoothing algorithm adopts the prior art, and is not described herein again.
In this embodiment, in step S2, an initialized LSTM model is constructed, that is, an initial shape is constructedA state LSTM neural network, which is a network structure formed by connecting a plurality of neural network cells, as shown in FIG. 2, wherein Ct-1Is the last cell state, ht-1For the last cell layer output, xtFor the input of the current cell, respectively using σf、σiAnd σoTo represent forgetting, input and output gates, tanh is set at the update gate and output gate of the LSTM neural network for generating the update content and updating the state of the neural network cells at the present moment, respectively.
In this embodiment, in step S3, the size of the sliding window for generating the sample is set, and the sliding window method is used to perform batch processing on the degradation characteristic parameters at different time points, that is, to convert the time series problem into supervised learning, and construct training sample data conforming to the input format of the LSTM model, and construct a sample sequence, as shown in fig. 3. Of course, in order to test the LSTM model trained in the later stage, test sample data can be obtained by the above method, and test verification can be performed on the trained prediction model (BO-LSTM) by using the test sample data as a test set.
In this embodiment, in step S4, adjusting the target parameter value in the LSTM model in a Bayesian Optimization (BO) manner to minimize the remaining life loss value output by the LSTM model specifically includes:
s41, setting the maximum optimization times N, randomly initializing target parameters within the range of the target parameters, taking the initialized target parameters as initial sampling points, and inputting the initial sampling points into Gaussian regression; wherein, the maximum optimization times N can be set according to experience;
s42, correcting the Gaussian regression according to the residual life loss value output by the LSTM model, so that the Gaussian regression is close to the distribution of a real function;
s43, selecting the next sampling point to be evaluated in the corrected Gaussian regression by using the acquisition function, inputting the sampling point xi as an input into an LSTM model for training, obtaining a new output yi of the target function, updating a set D { (x1, y1), (x2, y2), … (xt, yt) } and the Gaussian regression model, and realizing correction of the Gaussian regression again;
s44, judging whether the correction times reach the optimization times N, if not, repeating the steps S42-S43; and if so, outputting a sampling point which enables the residual life loss value to be minimum.
According to the method, the maximum optimization times can be set to be 60 times, and the dynamic optimization process of the bayesian optimization algorithm is shown by fig. 4:
the XYZ axes distribution represents three hyper-parameters to be optimized, which are Initial learning rate (Initial learning rate), Number of hidden layer layers (Number of hidden layer units) and Number of hidden layer units (Number of hidden layer units),
in fig. 4, circles and hexagons both represent sampling points for the bayesian optimization process, and they differ in that the circle fitting error is greater than 0.005, and the hexagons are less than 0.005; the dashed lines between them represent the optimal trajectory, and two dividing planes in the space divide the whole space into three subspaces S1, S2 and S3. Through statistics, the proportion of the distribution of the sampling points at S1, S2 and S3 is 2: 1: 1, the hexagons with smaller errors are clearly more densely distributed than the circles with larger errors, and their spatial distribution is also more concentrated at S1. And in all sampling points, the five-pointed star is the optimal sampling point and corresponds to the optimal hyper-parameter combination.
By the method, aiming at the complex optimization problems of unknown target function expressions and high search cost of the hyper-parameters of the deep learning model, a continuously updated probability model is used, the posterior probability of the optimization function is updated through few target function evaluations, the optimal model hyper-parameter combination is obtained, and the method is suitable for the hyper-parameter tuning problem of the service life prediction model of the switching device. The purpose of BO hyper-parameters is to get an optimal hyper-parameter combination by iteration. The hyperparametric combination selection can be expressed as:
wherein f (x) is a minimized objective function for evaluating the optimal performance of the objective function; x is a radical of a fluorine atom*For the finally obtained optimal super parameter setAnd (6) mixing. The super-parameter is an adjusting parameter, namely a target parameter of the invention.
In this embodiment, in order to ensure that the LSTM model training can be smoothly performed and prevent the model training from being stuck, step S41 further includes: setting a maximum optimization time T;
in step S44, the method further includes: judging whether the execution time of the steps S42-S43 is greater than the optimization time T, if not, repeating the steps S42-S43; and if so, outputting a sampling point which enables the residual life loss value to be minimum.
In this embodiment, the probability distribution of the optimization process is determined according to the following formula:
wherein p (E | D) is the posterior probability,is the probability coefficient, p (E) is the prior probability; e is sampling point distribution, and D is acquisition function evaluation;
the prediction accuracy of the method for predicting the service life of the switching device is verified and evaluated as follows:
selecting Mean Absolute Error (MAE), Root Mean Square Error (RMSE), goodness of fit (R)2) To measure the prediction accuracy. Wherein, the MAE directly reflects the error magnitude, the RMSE reflects the overall prediction accuracy, R2Reflecting the model's ability to interpret the data. The smaller the MAE and RMSE, the more accurate the prediction, R2The closer to 1 the value of (a), the more accurate the prediction result. The calculation formula of the three is as follows:
wherein, yiA value corresponding to the true remaining life;predicting a residual life corresponding value for the prediction model;the average value of the corresponding values of the real residual life; m is the number of samples in the test set;
the Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and goodness of fit (R) of the prediction models2) The performance indexes below are shown in FIGS. 5 to 7, respectively. Wherein, the prediction model of the invention is BO-LSTM; the predicted effects of different prediction models are compared, as shown in table 1:
TABLE 1
Fig. 5-7 and table 1 all show that the prediction method of the switching device of the present invention has high accuracy and good effect.
Finally, the above embodiments are only for illustrating the technical solutions of the present invention and not for limiting, although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications or equivalent substitutions may be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all of them should be covered in the claims of the present invention.
Claims (7)
1. A switching device service life prediction method based on degradation characteristics is characterized in that: the method comprises the following steps:
s1, collecting degradation characteristic parameter information of a switch device;
s2, constructing an initialized LSTM model;
s3, making the degeneration characteristic parameter information into training sample data conforming to an input format of an LSTM model;
s4, inputting training sample data into the LSTM model, adjusting target parameter values in the LSTM model to enable the residual life loss value output by the LSTM model to be minimum, and taking the LSTM model set when the loss value is minimum as the well-trained LSTM model; the target parameters comprise the number of hidden layer layers, the number of hidden layer units and an initial learning rate;
and S5, inputting the degradation characteristic parameter information of the switching device to be tested into the well-trained LSTM model, and outputting the predicted value of the residual life of the switching device to be tested.
2. The degradation-feature-based switching device life prediction method of claim 1, wherein: the step S1 further includes: and carrying out noise reduction processing on the degradation characteristic parameter information to obtain the degradation characteristic parameter information after noise reduction.
3. The degradation feature-based switching device life prediction method of claim 1, wherein: adjusting the target parameter value in the LSTM model to minimize the remaining life loss value output by the LSTM model, specifically including:
s41, setting the maximum optimization times N, randomly initializing target parameters, taking the initialized target parameters as initial sampling points, and inputting the initial sampling points into Gaussian regression;
s42, correcting the Gaussian regression according to the residual life loss value output by the LSTM model, so that the Gaussian regression is close to the distribution of a real function;
s43, selecting a next sampling point to be evaluated from the corrected Gaussian regression by using an acquisition function, transmitting the sampling point serving as an input into an LSTM model for training, and correcting the Gaussian regression again;
s44, judging whether the correction times reach the optimization times N, if not, repeating the steps S42-S43; and if so, outputting a sampling point which enables the residual life loss value to be minimum.
4. The degradation-feature-based switching device life prediction method of claim 3, wherein: in step S41, the method further includes: setting a maximum optimization time T;
in step S44, the method further includes: judging whether the execution time of the steps S42-S43 is greater than the optimization time T, if not, repeating the steps S42-S43; and if so, outputting a sampling point which enables the residual life loss value to be minimum.
5. The degradation-feature-based switching device life prediction method of claim 3, wherein: determining the probability distribution of the optimization process according to the following formula:
6. The degradation-feature-based switching device life prediction method of claim 1, wherein: and manufacturing training sample data by adopting a sliding window method.
7. The degradation-feature-based switching device life prediction method of claim 1, wherein: the loss value is calculated using the root mean square difference.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210255617.2A CN114611803A (en) | 2022-03-15 | 2022-03-15 | Switch device service life prediction method based on degradation characteristics |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210255617.2A CN114611803A (en) | 2022-03-15 | 2022-03-15 | Switch device service life prediction method based on degradation characteristics |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114611803A true CN114611803A (en) | 2022-06-10 |
Family
ID=81863551
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210255617.2A Pending CN114611803A (en) | 2022-03-15 | 2022-03-15 | Switch device service life prediction method based on degradation characteristics |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114611803A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2024045351A1 (en) * | 2022-08-29 | 2024-03-07 | 北京智芯微电子科技有限公司 | Method and apparatus for predicting service life of cmos device, electronic device, and medium |
-
2022
- 2022-03-15 CN CN202210255617.2A patent/CN114611803A/en active Pending
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2024045351A1 (en) * | 2022-08-29 | 2024-03-07 | 北京智芯微电子科技有限公司 | Method and apparatus for predicting service life of cmos device, electronic device, and medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108304623B (en) | Probability load flow online calculation method based on stack noise reduction automatic encoder | |
CN109991542B (en) | Lithium ion battery residual life prediction method based on WDE optimization LSTM network | |
CN107992976B (en) | Hot topic early development trend prediction system and prediction method | |
CN111860982A (en) | Wind power plant short-term wind power prediction method based on VMD-FCM-GRU | |
CN110059867B (en) | Wind speed prediction method combining SWLSTM and GPR | |
CN111563706A (en) | Multivariable logistics freight volume prediction method based on LSTM network | |
CN112988723A (en) | Traffic data restoration method based on space self-attention-diagram convolution cyclic neural network | |
CN113537600B (en) | Medium-long-term precipitation prediction modeling method for whole-process coupling machine learning | |
CN113705877B (en) | Real-time moon runoff forecasting method based on deep learning model | |
CN109063939A (en) | A kind of wind speed forecasting method and system based on neighborhood door shot and long term memory network | |
CN111861013A (en) | Power load prediction method and device | |
CN111612262A (en) | Wind power probability prediction method based on quantile regression | |
CN115275991A (en) | Active power distribution network operation situation prediction method based on IEMD-TA-LSTM model | |
CN110633859A (en) | Hydrological sequence prediction method for two-stage decomposition integration | |
CN113962145A (en) | Parameter uncertainty quantitative modeling method under interval data sample condition | |
CN115114128A (en) | Satellite health state evaluation system and evaluation method | |
CN113537469A (en) | Urban water demand prediction method based on LSTM network and Attention mechanism | |
CN115308558A (en) | Method and device for predicting service life of CMOS (complementary Metal oxide semiconductor) device, electronic equipment and medium | |
CN114611803A (en) | Switch device service life prediction method based on degradation characteristics | |
CN114757441A (en) | Load prediction method and related device | |
CN112363140A (en) | Thermodynamic constraint extrapolation objective correction method based on cyclic neural network | |
WO2024087129A1 (en) | Generative adversarial multi-head attention neural network self-learning method for aero-engine data reconstruction | |
CN115358178B (en) | Circuit yield analysis method based on fusion neural network | |
CN115794805A (en) | Medium-low voltage distribution network measurement data supplementing method | |
CN115796327A (en) | Wind power interval prediction method based on VMD (vertical vector decomposition) and IWOA-F-GRU (empirical mode decomposition) -based models |
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 |