CN117250490A - LSTM-based residual current operated circuit breaker service life prediction method - Google Patents
LSTM-based residual current operated circuit breaker service life prediction method Download PDFInfo
- Publication number
- CN117250490A CN117250490A CN202311213185.XA CN202311213185A CN117250490A CN 117250490 A CN117250490 A CN 117250490A CN 202311213185 A CN202311213185 A CN 202311213185A CN 117250490 A CN117250490 A CN 117250490A
- Authority
- CN
- China
- Prior art keywords
- neural network
- lstm
- circuit breaker
- residual current
- operated circuit
- 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 65
- 238000013528 artificial neural network Methods 0.000 claims abstract description 85
- 238000012360 testing method Methods 0.000 claims abstract description 64
- 238000012549 training Methods 0.000 claims abstract description 51
- 238000006731 degradation reaction Methods 0.000 claims abstract description 39
- 230000015556 catabolic process Effects 0.000 claims abstract description 38
- 230000001133 acceleration Effects 0.000 claims abstract description 31
- 238000005070 sampling Methods 0.000 claims abstract description 25
- 230000009471 action Effects 0.000 claims abstract description 24
- 230000006870 function Effects 0.000 claims description 75
- 230000008569 process Effects 0.000 claims description 23
- 230000014509 gene expression Effects 0.000 claims description 14
- 238000004364 calculation method Methods 0.000 claims description 13
- 230000000694 effects Effects 0.000 claims description 8
- 210000002569 neuron Anatomy 0.000 claims description 8
- 239000003795 chemical substances by application Substances 0.000 claims description 6
- 238000011156 evaluation Methods 0.000 claims description 6
- 230000001105 regulatory effect Effects 0.000 claims description 6
- 238000005315 distribution function Methods 0.000 claims description 3
- 238000012545 processing Methods 0.000 abstract description 4
- 239000010410 layer Substances 0.000 description 32
- 230000035882 stress Effects 0.000 description 18
- 210000004027 cell Anatomy 0.000 description 7
- 238000005457 optimization Methods 0.000 description 5
- 101001095088 Homo sapiens Melanoma antigen preferentially expressed in tumors Proteins 0.000 description 2
- 102100037020 Melanoma antigen preferentially expressed in tumors Human genes 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 2
- 238000013459 approach Methods 0.000 description 2
- 230000007774 longterm Effects 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000003213 activating effect Effects 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 230000032683 aging Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000001276 controlling effect Effects 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 238000013213 extrapolation Methods 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
- 239000002356 single layer Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/327—Testing of circuit interrupters, switches or circuit-breakers
-
- 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
- G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to the technical field of life prediction of electronic components and provides a life prediction method of a residual current operated circuit breaker based on LSTM. The method comprises the following steps: setting up an acceleration test platform, carrying out an acceleration degradation test on the residual current operated circuit breaker to obtain degradation data, and dividing the degradation data into a training set and a testing set; constructing a first neural network based on LSTM by taking a training set as an input vector, wherein the training set is a residual action current degradation track; defining a neural network parameter of a first neural network, fitting an objective function through a probability proxy model, and optimizing parameter points of the neural network parameter through a sampling function; training the first neural network according to the optimized neural network parameters to obtain a second neural network; and inputting the test set into a second neural network to obtain a life prediction result of the residual current operated circuit breaker. The invention improves the prediction precision and the precision of processing time sequence data, and has stronger stability.
Description
Technical Field
The invention relates to the technical field of life prediction of electronic components, in particular to a life prediction method of a residual current operated circuit breaker based on LSTM.
Background
The residual current operated circuit breaker is a key electrical appliance for protecting personnel safety of electricity consumption personnel and preventing leakage accidents in a power system. A high reliability is required for the residual current operated circuit breaker. Over time, the performance of the residual current operated circuit breaker gradually deteriorates due to the aging of the internal electronic devices, and finally the circuit breaker cannot work normally, so that huge potential safety hazards are brought to the users. Therefore, the life prediction of the residual current operated circuit breaker is significant in guaranteeing the reliable operation of the power system.
For a residual current operated circuit breaker, a life prediction method based on failure physics needs to respectively study degradation mechanisms of components in a product, abstract mathematical expressions of product degradation, is complex to realize, and modeling accuracy is difficult to guarantee; in addition, the lifetime prediction method of the degradation track modeling needs to obtain a mathematical expression of the degradation track through fitting, and in terms of operability, the pseudo failure lifetime data is obtained by extrapolation through linear fitting, but the degradation process of an actual product has certain randomness, the degradation tracks of different products are different, and the linear fitting method only has strong limitation and poor universality. Meanwhile, the degradation track of the actual product does not strictly follow the linear rule, and the method adopting track fitting has larger error.
Disclosure of Invention
The present invention is directed to solving at least one of the technical problems existing in the related art. Therefore, the invention provides a residual current operated circuit breaker life prediction method based on LSTM.
The invention provides a residual current operated circuit breaker life prediction method based on LSTM, which comprises the following steps:
s1: setting up an acceleration test platform, carrying out an acceleration degradation test on a residual current operated circuit breaker, obtaining degradation data, and dividing the degradation data into a training set and a testing set;
s2: constructing a first neural network based on LSTM by taking the training set as an input vector, wherein the training set is a residual action current degradation track in the degradation data;
s3: defining a neural network parameter of the first neural network, fitting an objective function through a probability proxy model, and optimizing parameter points of the neural network parameter through a sampling function so as to optimize the neural network parameter;
s4: training the first neural network according to the optimized neural network parameters to obtain a second neural network;
s5: and inputting the test set into the second neural network to obtain a life prediction result of the residual current operated circuit breaker.
According to the LSTM-based residual current operated circuit breaker service life prediction method provided by the invention, in the step S1, the acceleration test platform provides acceleration stress through a temperature and humidity regulating box.
According to the residual current operated circuit breaker life prediction method based on LSTM provided by the invention, the step S2 further comprises the following steps:
and normalizing the input vector by a MAX-MIN method.
According to the residual current operated circuit breaker life prediction method based on LSTM provided by the invention, the first neural network training process in the step S4 comprises forward calculation model output, error term back propagation and gradient calculation according to the error term.
According to the residual current operated circuit breaker life prediction method based on LSTM provided by the invention, in step S3, the probability agent model completes fitting objective function through Gaussian process, and the Gaussian distribution expression of the probability agent model is:
wherein,for an objective function obtained by increasing the dimension of the gaussian distribution, +.>Is Gaussian in shapeProcedure (S)/(S)>Desired for the objective function->As a covariance function.
According to the residual current operated circuit breaker life prediction method based on LSTM provided by the invention, the expression of the sampling function in the step S3 is as follows:
wherein,for sampling function +.>Desired for sampling function +.>For the current optimum value in the process of optimizing the sampling function,/for the sampling function>Is standard deviation (S)>For the desired degree of elevation, add>Distribution function of standard normal distribution, +.>Is a probability density function of a standard normal distribution.
According to the residual current operated circuit breaker service life prediction method based on LSTM provided by the invention, the posterior distribution expression of the function corresponding to the optimized neural network parameter in the step S4 is as follows:
wherein,for posterior distribution of functions corresponding to the optimized neural network parameters, ++>Likelihood distribution of functions corresponding to the optimized neural network parameters +.>For the a priori distribution of the functions corresponding to the optimized neural network parameters, ++>And (3) marginal likelihood distribution of functions corresponding to the optimized neural network parameters.
According to the LSTM-based residual current operated circuit breaker service life prediction method provided by the invention, the first neural network built in the step S2 takes root mean square error as a training effect evaluation index, and the expression of the root mean square error is as follows:
wherein,is root mean square error>For training data number +.>For index value->To output the true value, +.>To output a predicted value.
According to the method for predicting the service life of the residual current operated circuit breaker based on the LSTM, the neural network parameters of the first neural network in the step S3 comprise the learning rate, the LSTM layer number and the LSTM neuron number.
The invention provides a residual current operated circuit breaker life prediction method based on LSTM, which optimizes the super parameters of the LSTM neural network, solves the problem of model super parameter selection, effectively improves model prediction precision, has higher precision when processing time sequence data, has good performance in long-term prediction of a short training set, has stronger stability, has stronger time sequence for a residual action current data sequence, can introduce historical data information, has higher prediction precision and stronger stability in time sequence prediction, and can be better applied to the prediction of the residual action current data.
Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention.
Drawings
In order to more clearly illustrate the invention or the technical solutions of the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described, and it is obvious that the drawings in the description below are some embodiments of the invention, and other drawings can be obtained according to these drawings without inventive effort for a person skilled in the art.
Fig. 1 is a flowchart of a residual current operated circuit breaker life prediction method based on LSTM according to an embodiment of the present invention.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the present invention more apparent, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings, and it is apparent that the described embodiments are some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention. The following examples are illustrative of the invention but are not intended to limit the scope of the invention.
In the description of the embodiments of the present invention, it should be noted that the terms "center", "longitudinal", "lateral", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. indicate orientations or positional relationships based on the orientations or positional relationships shown in the drawings, are merely for convenience in describing the embodiments of the present invention and simplifying the description, and do not indicate or imply that the apparatus or elements referred to must have a specific orientation, be configured and operated in a specific orientation, and thus should not be construed as limiting the embodiments of the present invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance.
In describing embodiments of the present invention, it should be noted that, unless explicitly stated and limited otherwise, the terms "coupled," "coupled," and "connected" should be construed broadly, and may be either a fixed connection, a removable connection, or an integral connection, for example; can be mechanically or electrically connected; can be directly connected or indirectly connected through an intermediate medium. The specific meaning of the above terms in embodiments of the present invention will be understood in detail by those of ordinary skill in the art.
In embodiments of the invention, unless expressly specified and limited otherwise, a first feature "up" or "down" on a second feature may be that the first and second features are in direct contact, or that the first and second features are in indirect contact via an intervening medium. Moreover, a first feature being "above," "over" and "on" a second feature may be a first feature being directly above or obliquely above the second feature, or simply indicating that the first feature is level higher than the second feature. The first feature being "under", "below" and "beneath" the second feature may be the first feature being directly under or obliquely below the second feature, or simply indicating that the first feature is less level than the second feature.
In the description of the present specification, a description referring to terms "one embodiment," "some embodiments," "examples," "specific examples," or "some examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the embodiments of the present invention. In this specification, schematic representations of the above terms are not necessarily directed to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, the different embodiments or examples described in this specification and the features of the different embodiments or examples may be combined and combined by those skilled in the art without contradiction.
An embodiment of the present invention is described below with reference to fig. 1.
The invention provides a residual current operated circuit breaker life prediction method based on LSTM, which comprises the following steps:
s1: setting up an acceleration test platform, carrying out an acceleration degradation test on a residual current operated circuit breaker, obtaining degradation data, and dividing the degradation data into a training set and a testing set;
the acceleration test platform in the step S1 provides acceleration stress through a temperature and humidity regulating box.
Further, the specific mode of acquiring the residual action current degradation data is to firstly use a temperature and humidity regulating box to provide acceleration stress required by an acceleration test, and use a residual current action characteristic measuring instrument to measure the residual action current value of a test sample.
S2: constructing a first neural network based on LSTM by taking the training set as an input vector, wherein the training set is a residual action current degradation track in the degradation data;
wherein, step S2 further comprises:
and normalizing the input vector by a MAX-MIN method.
The first neural network constructed in the step S2 takes root mean square error as a training effect evaluation index, and the expression of the root mean square error is as follows:
wherein,is root mean square error>For training data number +.>For index value->To output the true value, +.>To output a predicted value.
Further, the modeling process of the LSTM neural network, that is, the first neural network, includes: defining an LSTM neural network structure; giving input data; dividing an input data sequence into a training set and a testing set; defining LSTM neural network parameters; and selecting model training effect evaluation indexes.
Furthermore, the first neural network constructed based on LSTM can increase or remove the information data in the memory unit by introducing a gate structure; the gate structure contains a sigmoid activation function whose value range is (0, 1), by which the amount of information data passing through the gate can be controlled, when the function is 0, no data information is allowed to pass, when the function value is 1, the information data can pass completely, and the gate structure contains three gates, namely a forgetting gate, an input gate and an output gate.
When calculating the state at the current moment, controlling the introduction quantity of the unit state at the last moment by the forgetting gate of the LSTM neural network; the input gate determines the input quantity input at the moment; the output gate controls how much of the output at that time can be used as the output value of the LSTM neural network; each complete section of the LSTM neural network is called a memory cell, each memory cell having three inputs of data: the output value at the previous time, the input at the current time and the state of the neuron cells at the previous time have two output data: neuron cell status and output value at the current time.
S3: defining a neural network parameter of the first neural network, fitting an objective function through a probability proxy model, and optimizing parameter points of the neural network parameter through a sampling function so as to optimize the neural network parameter;
in step S3, the probability agent model completes fitting the objective function through a gaussian process, where the gaussian distribution expression of the probability agent model is:
wherein,for an objective function obtained by increasing the dimension of the gaussian distribution, +.>For Gaussian process, < >>Desired for the objective function->As a covariance function.
Further, a probabilistic proxy model is employed to fit the objective function. Along with the increasing of the information acquired in the optimizing process, the prior distribution of the proxy model is continuously corrected, so that the proxy model continuously approaches the objective function. This requires that the proxy model be able to build a distribution arbitrarily in the objective function, typically by choosing a gaussian process to accomplish this, by increasing the dimension of the gaussian distribution, an arbitrary objective function can be approximated.
The expression of the sampling function in step S3 is:
wherein,for sampling function +.>Desired for sampling function +.>For the current optimum value in the process of optimizing the sampling function,/for the sampling function>Is standard deviation (S)>For the desired degree of elevation, add>Distribution function of standard normal distribution, +.>Is a probability density function of a standard normal distribution.
Further, the sampling function is used to continuously find the function point which is the best possible. The sampling function searches the next possible optimal function point based on the known history information, the process is called exploration, when the sampling function finds the local optimal solution, the value is continuously taken at the position, the process is called utilization, and the function optimal solution can be obtained through iteration by balancing the utilization process and the exploration process.
The sampling function can show the lifting degree of the next function point in the optimizing process in the expected aspect compared with the current function point, so that the point with the largest lifting is selected as the next function point.
The neural network parameters of the first neural network in step S3 include a learning rate, an LSTM layer number, and an LSTM neuron number.
S4: training the first neural network according to the optimized neural network parameters to obtain a second neural network;
wherein, the first neural network training process in step S4 includes forward calculation of model output, error term back propagation, and gradient calculation according to the error term.
Further, forward propagation computation mainly includes computation of forget gate, input gate, output gate, memory cell and final output.
The calculation formula of the forgetting gate output is as follows:
wherein,output for forgetting gate, ++>To activate the function +.>For forgetting the gate weight, +.>Implicit layer state of memory cell at last moment, < >>For the current time input, < >>Is a forget gate threshold.
The input gate output is calculated as follows:
wherein,for input gate output, +.>For inputting gate weight->Is the input gate threshold.
The calculation formula of the memory cell state is as follows:
wherein,is->Time memory state unit->Indicating that new information is added to the memory cell,/for the memory cell>Is->Weight of time state quantity +.>Is->Threshold value of time state quantity.
The output gate output is calculated as follows:
wherein,for outputting gate output, +.>For outputting the gate weight +.>Is the output gate threshold.
The final output value calculation formula is as follows:
the state quantity of the memory unit is multiplied by the output gate output after being processed by the tangent function, and a final output value is obtained.
Error back propagation involves two aspects: the error term is defined as the partial derivative of the loss function with respect to the output value along time and layer counter-propagation, and is calculated as follows:
wherein,is error item->For loss function->Is a partial derivative operation.
Activating the back propagation of the error term, taking the forgetting gate as an example, calculating the weighted input of the moment and the error term, and the other gates are similar in calculation process, and the calculation formula is as follows:
wherein,and (5) inputting the weight corresponding to the forget gate.
Then, first willThe error term of the moment is counter-propagated in time, giving +.>The error term of the moment is then propagated back along the layer, assuming that from +.>Layer propagation to->Each time an error propagates along time and in the opposite direction of the layer, the error is multiplied by the corresponding error term.
The total number of the gradients of the weight and the threshold value to be calculated is 12, and the final gradients of the input gate, the forgetting gate, the output gate and the weight and the threshold value corresponding to the states of the memory unit are equal to the sum of the gradients of all the moments, so that the updating of the weight threshold value of each gate at all the moments can be realized.
The posterior distribution expression of the function corresponding to the optimized neural network parameter in step S4 is:
wherein,for posterior distribution of functions corresponding to the optimized neural network parameters, ++>Likelihood distribution of functions corresponding to the optimized neural network parameters +.>For the a priori distribution of the functions corresponding to the optimized neural network parameters, ++>And (3) marginal likelihood distribution of functions corresponding to the optimized neural network parameters.
S5: and inputting the test set into the second neural network to obtain a life prediction result of the residual current operated circuit breaker.
In some embodiments, an acceleration test scheme of the residual current operated circuit breaker is formulated, an acceleration test platform is built, and an acceleration degradation test is performed on the residual current operated circuit breaker to obtain residual operation current degradation data.
Firstly, an acceleration test scheme of a residual current operated circuit breaker is established, the test type is a constant stress acceleration degradation test, the acceleration stress is temperature, the residual operation current is selected as a performance degradation characteristic quantity, four stress levels are set to be 55 ℃, 65 ℃,77 ℃, 90 ℃, test periods of 55 ℃ and 65 ℃ are 200 days, test periods of 77 ℃ and 90 ℃ are from test to sample failure, each period is 24 hours, and the number of test samples is 4 under each stress level.
An acceleration test platform is built and mainly comprises a residual current action characteristic measuring instrument, a temperature and humidity regulating box, a residual current action circuit breaker sample and an upper computer. The residual current action characteristic measuring instrument is used for measuring the residual action current value of the test sample, the temperature and humidity regulating box provides acceleration stress required by the acceleration test, and the upper computer records and analyzes data.
The method mainly comprises five processes of test sample selection, initial residual action current test, accelerated degradation circulation, residual action current test and test termination judgment.
And (3) completing four groups of residual current operated circuit breakers acceleration tests according to the test flow, and carrying out drawing analysis on test degradation data to obtain the degradation tracks of residual operation currents of the test sample under different temperature stress levels.
And secondly, carrying out degradation track modeling by adopting an LSTM neural network, wherein the basic modeling steps are as follows.
Definition of LSTM neural network structure: the constructed LSTM neural network prediction model consists of an input layer, an LSTM layer, a Dropout layer, a full connection layer and an output layer. The input layer receives input data, performs normalization processing on the data, removes dimensions and reduces the calculation time of the neural network, and the input layer is set to be 1 layer; the LSTM layer carries out learning training on the data transferred by the input layer, and according to research, the single-layer neural network can process various regression problems, so that the LSTM layer is defined as 1 layer, and the number of hidden units is set to 128; the Dropout layer can set the input data to zero with a certain probability, the operation can effectively reduce network overfitting, and the probability weight of the Dropout layer is set to 0.5; the full-connection layer can collect the transmitted information data, process the information by adopting the weight and the threshold value, and set the weight of the layer as 1; the output layer processes the transferred data and outputs the processed data, and the output layer is set to be 1.
Given input data: the input vector is the residual action current degradation track obtained by the acceleration test, the input vector is normalized by adopting a MAX-MIN method, sliding window prediction is adopted, the first 8 data are input as a model, the 9 th data are output as a model, and every 9 data are used as a training sample.
Dividing the input data sequence into a training set and a testing set: the training set is used for training an LSTM neural network prediction model, the testing set is used for verifying the effect of model prediction, when the output of the training sample corresponds to the last data of the training set, model training is completed, and the prediction model established by the LSTM neural network is as follows:
defining LSTM neural network parameters: the model gradient function optimization method adopts a self-adaptive moment estimation method, the method is stable for the estimated value of the model parameter, and the iteration step length is set to be 1; the maximum iteration number of the model is 250; when the model learning rate is function optimizing, the step size multiplied by the gradient can control the training speed of the model, the model training can be too slow due to too small value, the model can vibrate nearby the optimal value due to too large value, the optimal value can not be obtained, therefore, the sectional learning rate is adopted, the learning rate is set to be 0.005 in the first 125 parameter iterations, the learning rate is multiplied by the coefficient of 0.2 in the last 125 iterations, and the learning rate is reduced to be 0.001.
Selecting model training effect evaluation indexes: the model training result evaluation index selects root mean square error (Root mean square error, RMSE).
During prediction, two groups are set for each data sequence, and different training and prediction set dividing modes are adopted for the two groups to test the prediction effect of the model. The training set and the test set of the group 1 respectively account for about 80% and 20% of the total data; the training set and the test set of group 2 each account for about 50% of the total.
Again, a probabilistic proxy model is employed to fit the objective function and the sampling function is used to continually find the function points that are potentially optimal.
Based on the principle described above, the super parameters of the LSTM neural network are optimized, an iterative optimization mode is adopted, the maximum iteration number is selected for 40 times, and the parameters to be optimized are the LSTM layer number, the LSTM layer neuron number and the learning rate; the LSTM layer optimizing range is an integer between [1,2], the LSTM layer neuron optimizing range is an integer between [30,200], and the learning rate optimizing range is [0.001,0.5].
From the experimental results, as the iteration times increase, the sampling function continuously obtains the target value with smaller error, the estimated value of the proxy function gradually approaches to the optimal target value, and after 40 iterations, the model parameters of the A1 sample group 2 obtained by Bayesian optimization are as follows: the LSTM layer number is 1 layer, the LSTM layer neuron number is 175, the initial learning rate is 0.0363, bayesian parameter optimization is respectively carried out on other groups of data, the LSTM neural network model is given with optimal super parameters, data prediction is carried out, and the prediction results and absolute errors of the prediction results of the two groups of sample residual action current sequence test sets can be obtained.
And carrying out error analysis on the prediction result of the LSTM neural network test set, and calculating average absolute error MAE, mean square error MSE, average absolute percentage error MAPE and maximum absolute percentage error APE_MAX between the prediction value and the actual measurement value of the LSTM neural network model, wherein the average absolute error MAE, the mean square error MSE, the average absolute percentage error MAPE and the maximum absolute percentage error APE_MAX are shown in table 1.
TABLE 1 prediction errors for LSTM neural networks
As can be seen from Table 1, the A1 and B1 sample data sequences have the same length, the same division mode for the training test set is also the same, and the prediction accuracy of the A1 sample data sequence is relatively high. Analyzing the reason, the test set of the residual action current data sequence of the B1 sample has a larger trend change in the last period, and the difficulty of LSTM model prediction is increased, so that the different degradation trend characteristics of the data sequence can influence the prediction result of the model.
Ape_max was 0.27% and 0.35% respectively when the training set accounted for 80% of the total data length and the test set accounted for 20%. When the training set and the test set respectively account for 50% of the total data length, the APE_MAX is respectively 0.50% and 1.09%, so that when the training set data are more sufficient, the LSTM neural network can obtain higher prediction precision; when the training set is less, the prediction error is only increased by 2-3 times of the original prediction error, and the prediction precision is stable.
The residual current operated circuit breaker does not run under the acceleration stress of the test in the actual situation, so that the service life of the residual current operated circuit breaker under the normal stress needs to be predicted, the service life of the residual current operated circuit breaker under the normal stress is extrapolated through an Arrhenius acceleration equation, and the failure life of the residual current operated circuit breaker under 4 groups of acceleration stress needs to be obtained. Through the analysis, the Bayesian optimization LSTM model has higher precision and stronger stability in the aspect of track prediction of the residual action current degradation data, so that the model is adopted to predict the pseudo failure life of the residual current action circuit breaker sample at 55 and 65 ℃.
And taking degradation data of 200 cycles at 55 and 65 ℃ as a training set, and predicting after training is finished to obtain the pseudo failure life of the residual current operated circuit breaker sample at 55 and 65 ℃. And adopting sliding window prediction, and filling the predicted value of each time into the sliding window until the residual action current of the residual current action circuit breaker reaches 15mA or less. The predicted pseudo-failure life of the test article was recorded, and the pseudo-failure life of each sample under temperature stress at 55 and 65 ℃ is shown in table 2.
TABLE 2 pseudo-failure life of each test sample
The average value of the pseudo failure life of each group was taken as the failure life of the sample under the temperature stress, and the same operation was performed on the failure life of the residual current operated circuit breaker obtained by the acceleration test at 70 and 85 ℃ to calculate the results shown in table 3 below.
Table 3 test article failure life under four test stress groups
After the failure life of the product under each high-temperature stress is calculated, the relation between the temperature stress and the product life is required to be established, the failure life of the residual current operated circuit breaker under the high-temperature test is extrapolated to the service life at normal temperature, the relation is established by adopting an Arrhenius model, and the logarithmic relation between the life characteristic quantity of the product and the reciprocal of the temperature is obtained.
According to the functional relation between the service life and the test temperature, the service life in the table 3 is substituted into the relation established by the Arrhenius model, the linear relation between the service life and the test temperature is fitted by adopting a least square method, and the slope of a service life regression curve is 5485.7 and the intercept is-10.58.
The linear fitting correlation coefficient is 0.9594, the value of the residual square sum RSS is 0.01939, the correlation coefficient is close to 1, the residual square sum RSS is close to 0, the fitting degree between the temperature and the service life is good, the linear regression effect is obvious, the normal temperature 25 ℃ is converted into the Kelvin temperature for calculation, and the service life of the residual current operated circuit breaker at the normal temperature can be extrapolated to be about 2479 days.
According to the life prediction method adopting the neural network, the life prediction value of the residual current operated circuit breaker is obtained, according to the existing research, when the residual current operated circuit breaker works to the time, the residual current operated circuit breaker is not damaged immediately, but is in a false operation fault hiding state, and only when the leakage current in a power supply circuit exceeds the current residual operation current value, the residual current operated circuit breaker can trip by false operation, but at the moment, the reliability of a product is greatly reduced, the probability of abnormal operation is higher, the operation characteristics of the product are detected, the abnormal operation is found out to be replaced in time, and the safety of a power distribution system is improved.
The invention provides a residual current operated circuit breaker life prediction method based on LSTM, which optimizes the super parameters of the LSTM neural network, solves the problem of model super parameter selection, effectively improves model prediction precision, has higher precision when processing time sequence data, has good performance in long-term prediction of a short training set, has stronger stability, has stronger time sequence for a residual action current data sequence, can introduce historical data information, has higher prediction precision and stronger stability in time sequence prediction, and can be better applied to the prediction of the residual action current data.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims (9)
1. The life prediction method of the residual current operated circuit breaker based on the LSTM is characterized by comprising the following steps of:
s1: setting up an acceleration test platform, carrying out an acceleration degradation test on a residual current operated circuit breaker, obtaining degradation data, and dividing the degradation data into a training set and a testing set;
s2: constructing a first neural network based on LSTM by taking the training set as an input vector, wherein the training set is a residual action current degradation track in the degradation data;
s3: defining a neural network parameter of the first neural network, fitting an objective function through a probability proxy model, and optimizing parameter points of the neural network parameter through a sampling function so as to optimize the neural network parameter;
s4: training the first neural network according to the optimized neural network parameters to obtain a second neural network;
s5: and inputting the test set into the second neural network to obtain a life prediction result of the residual current operated circuit breaker.
2. The LSTM-based residual current operated circuit breaker life prediction method according to claim 1, wherein the acceleration test platform in step S1 provides acceleration stress through a temperature and humidity regulating box.
3. The LSTM-based residual current operated circuit breaker life prediction method according to claim 1, wherein step S2 further includes:
and normalizing the input vector by a MAX-MIN method.
4. The LSTM based residual current operated circuit breaker life prediction method according to claim 1, wherein the first neural network training process in step S4 includes forward calculation of model output, error term back propagation and calculation of gradients from error term.
5. The method for predicting the life of the residual current operated circuit breaker based on the LSTM according to claim 1, wherein in the step S3, the probability agent model completes fitting an objective function through a gaussian process, and the expression of the gaussian distribution of the probability agent model is:
wherein,for an objective function obtained by increasing the dimension of the gaussian distribution, +.>Is Gaussian in shapeProcedure (S)/(S)>Desired for the objective function->As a covariance function.
6. The LSTM-based residual current operated circuit breaker life prediction method according to claim 1, wherein the expression of the sampling function in step S3 is:
wherein,for sampling function +.>Desired for sampling function +.>For the current optimum value in the process of optimizing the sampling function,/for the sampling function>Is standard deviation (S)>For the desired degree of elevation, add>Distribution function of standard normal distribution, +.>Is a probability density function of a standard normal distribution.
7. The LSTM-based residual current operated circuit breaker life prediction method according to claim 1, wherein the posterior distribution expression of the function corresponding to the optimized neural network parameter in step S4 is:
wherein,for posterior distribution of functions corresponding to the optimized neural network parameters, ++>Likelihood distribution of functions corresponding to the optimized neural network parameters +.>For the a priori distribution of the functions corresponding to the optimized neural network parameters, ++>And (3) marginal likelihood distribution of functions corresponding to the optimized neural network parameters.
8. The LSTM-based life prediction method of a residual current operated circuit breaker according to claim 1, wherein the first neural network built in step S2 uses a root mean square error as a training effect evaluation index, and the expression of the root mean square error is:
wherein,is root mean square error>For training data number +.>For index value->To output the true value, +.>To output a predicted value.
9. The method according to claim 1, wherein the neural network parameters of the first neural network in step S3 include learning rate, LSTM layer number and LSTM neuron number.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311213185.XA CN117250490A (en) | 2023-09-20 | 2023-09-20 | LSTM-based residual current operated circuit breaker service life prediction method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311213185.XA CN117250490A (en) | 2023-09-20 | 2023-09-20 | LSTM-based residual current operated circuit breaker service life prediction method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117250490A true CN117250490A (en) | 2023-12-19 |
Family
ID=89125868
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311213185.XA Pending CN117250490A (en) | 2023-09-20 | 2023-09-20 | LSTM-based residual current operated circuit breaker service life prediction method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117250490A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117877028A (en) * | 2024-03-13 | 2024-04-12 | 浙江大学 | Motor insulation life prediction method and system based on microscopic image features |
-
2023
- 2023-09-20 CN CN202311213185.XA patent/CN117250490A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117877028A (en) * | 2024-03-13 | 2024-04-12 | 浙江大学 | Motor insulation life prediction method and system based on microscopic image features |
| CN117877028B (en) * | 2024-03-13 | 2024-05-14 | 浙江大学 | Motor insulation life prediction method and system based on microscopic image features |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wang et al. | Improved anti-noise adaptive long short-term memory neural network modeling for the robust remaining useful life prediction of lithium-ion batteries | |
| CN112036084B (en) | Similar product life migration screening method and system | |
| Lin et al. | State of health estimation of lithium-ion battery based on an adaptive tunable hybrid radial basis function network | |
| CN109061504A (en) | Same type difference lithium ion battery remaining life prediction technique and system | |
| Chen et al. | Remaining useful life prediction of lithium-ion battery using a novel particle filter framework with grey neural network | |
| CN111965476B (en) | Low-voltage diagnosis method based on graph convolution neural network | |
| CN114325450A (en) | Lithium ion battery health state prediction method based on CNN-BilSTM-AT hybrid model | |
| Takyi-Aninakwa et al. | An optimized relevant long short-term memory-squared gain extended Kalman filter for the state of charge estimation of lithium-ion batteries | |
| Ouyang et al. | A novel state of charge estimation method for lithium-ion batteries based on bias compensation | |
| CN117250490A (en) | LSTM-based residual current operated circuit breaker service life prediction method | |
| Zhou et al. | Research on state of health prediction model for lithium batteries based on actual diverse data | |
| Li et al. | Data-driven state-of-health estimation for lithium-ion battery based on aging features | |
| CN113917334A (en) | Battery health state estimation method based on evolution LSTM self-encoder | |
| CN113761751A (en) | Lithium ion battery residual life prediction method and system based on temperature acceleration factor | |
| CN113850320A (en) | Transformer fault detection method based on improved support vector machine regression algorithm | |
| CN116930609A (en) | Electric energy metering error analysis method based on ResNet-LSTM model | |
| Li et al. | A hybrid framework for predicting the remaining useful life of battery using Gaussian process regression | |
| Ardeshiri et al. | Gated recurrent unit least-squares generative adversarial network for battery cycle life prediction | |
| Zhu et al. | A noise‐immune model identification method for lithium‐ion battery using two‐swarm cooperative particle swarm optimization algorithm based on adaptive dynamic sliding window | |
| Chen et al. | State of health estimation for lithium-ion battery based on particle swarm optimization algorithm and extreme learning machine | |
| Lv et al. | Analysis of strategies to maximize the cycle life of lithium-ion batteries based on aging trajectory prediction | |
| Tian et al. | Long-short term memory neural network based life prediction of lithium-ion battery considering internal parameters | |
| Zhang et al. | A method for remaining discharge time prediction of lithium‐ion batteries under dynamic uncertainty | |
| Guo et al. | Health prognostics of lithium-ion batteries based on universal voltage range features mining and adaptive multi-Gaussian process regression with Harris Hawks optimization algorithm | |
| Li et al. | 3-Dimensional general ADT modeling and analysis: Considering epistemic uncertainties in unit, time and stress dimension |
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 |