CN115906648A - Method and system for predicting remaining service life of transformer under severe variable working conditions - Google Patents
Method and system for predicting remaining service life of transformer under severe variable working conditions Download PDFInfo
- Publication number
- CN115906648A CN115906648A CN202211531192.XA CN202211531192A CN115906648A CN 115906648 A CN115906648 A CN 115906648A CN 202211531192 A CN202211531192 A CN 202211531192A CN 115906648 A CN115906648 A CN 115906648A
- Authority
- CN
- China
- Prior art keywords
- data
- service life
- transformer
- residual service
- neural network
- 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 34
- 238000012544 monitoring process Methods 0.000 claims abstract description 49
- 238000003062 neural network model Methods 0.000 claims abstract description 43
- 238000012549 training Methods 0.000 claims abstract description 34
- 125000004122 cyclic group Chemical group 0.000 claims abstract description 31
- 238000012360 testing method Methods 0.000 claims abstract description 25
- 238000005065 mining Methods 0.000 claims abstract description 14
- 230000000306 recurrent effect Effects 0.000 claims description 19
- 239000013598 vector Substances 0.000 claims description 16
- 238000005457 optimization Methods 0.000 claims description 8
- 230000003044 adaptive effect Effects 0.000 claims description 6
- 238000006243 chemical reaction Methods 0.000 claims description 5
- 238000013480 data collection Methods 0.000 claims description 5
- 238000010276 construction Methods 0.000 claims description 4
- 238000009412 basement excavation Methods 0.000 claims description 3
- 238000005516 engineering process Methods 0.000 claims description 2
- 238000013528 artificial neural network Methods 0.000 description 10
- 230000006870 function Effects 0.000 description 6
- 238000010586 diagram Methods 0.000 description 3
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 2
- 238000012417 linear regression Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- XEEYBQQBJWHFJM-UHFFFAOYSA-N Iron Chemical group [Fe] XEEYBQQBJWHFJM-UHFFFAOYSA-N 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a prediction method for the residual service life of a transformer under severe and variable working conditions, wherein a model training stage comprises the following steps: simultaneously inputting multi-sensing monitoring data and running state data into a multi-dimensional cyclic neural network model through different input channels, mining hidden features of different dimensions, and outputting predicted residual service life; adjusting and optimizing the multidimensional cyclic neural network model by using the actual residual service life and the output prediction deviation between the predicted residual service life; the prediction phase comprises: collecting test data of a large transformer to be predicted, wherein the test data comprises multi-sensor monitoring data and running state data; and inputting the test data into the trained multidimensional cyclic neural network model in real time to obtain the predicted residual service life of the large-scale transformer. By modeling and analyzing the multi-sensor monitoring data and the operation state data, the residual service life of the large transformer can be accurately predicted.
Description
Technical Field
The invention relates to the technical field of transformers, in particular to a method and a system for predicting the residual service life of a transformer under severe and variable working conditions.
Background
Sudden failure of a large transformer can result in significant economic loss. The residual service life prediction of the large transformer can be carried out to effectively estimate the residual effective working time of the large transformer under a certain performance level, and the method is an important means for ensuring the reliability, the availability, the maintainability and the safety of the large transformer.
The prediction of the residual service life of the large transformer under severe and variable working conditions is a very challenging problem, and not only multi-sensing monitoring data of the large transformer need to be analyzed, but also running state data need to be comprehensively utilized. Although part of existing methods for predicting the remaining service life of the large transformer claim that the methods can be used for predicting the remaining service life of the large transformer under variable working conditions, the methods usually only analyze monitoring data of the large transformer, and cannot realize simultaneous modeling analysis of the monitoring data and the operating state data, so that the methods are poor in prediction performance on the large transformer under most variable working conditions.
Disclosure of Invention
In view of the above, the invention aims to solve the problem that the prediction performance of the existing prediction method for the residual service life of the large-scale transformer is poor under most of variable working conditions.
In order to solve the technical problems, the invention provides the following technical scheme:
in a first aspect, the invention provides a method for predicting the remaining service life of a transformer under severe variable working conditions, which comprises the following steps: a model training phase and a prediction phase;
the model training phase comprises:
s11, simultaneously inputting multi-sensor monitoring data and running state data into a multi-dimensional recurrent neural network model through different input channels, mining hidden features of different dimensions, and outputting predicted residual service life;
s12, adjusting and optimizing the multidimensional cyclic neural network model by using the prediction deviation between the real residual service life and the output predicted residual service life;
the prediction phase comprises:
s21, collecting test data of the large transformer to be predicted, wherein the test data comprises multi-sensor monitoring data and running state data;
and S22, inputting the test data into the trained multi-dimensional cyclic neural network model in real time to obtain the predicted residual service life of the large transformer.
Further, step S11 includes:
s111, collecting historical data of the large transformer, and constructing a training data set, wherein the historical data comprises multi-sensor monitoring data, running state data and corresponding real remaining service life;
step S112, simultaneously inputting the multi-sensing monitoring data and the running state data in the training data set into the multi-dimensional recurrent neural network model through different input channels;
step S113, mining input data characteristics through a BLSTM layer and a BGRU layer which are parallel to each other so as to capture hidden characteristics from different dimensions;
and S114, outputting the predicted residual service life of the large transformer from the full connection layer according to the hidden features of different dimensions of excavation.
Further, step S112 includes:
step S1121, firstly performing linear conversion on the input multi-sensor monitoring data in two full-connection layers to obtain data hiding characteristics;
and step S1122, merging the running state data into the converted multi-sensing monitoring data to construct a high-order vector.
Further, in step S113, a dropout technique is employed to prevent repeated capture of the same feature.
Further, in step S12, the multidimensional recurrent neural network model is adjusted and optimized by using an adaptive moment estimation algorithm.
In a second aspect, the present invention provides a system for predicting remaining service life of a transformer under severe variable working conditions, including: the model training module and the prediction module;
the model training module comprises:
the model construction unit is used for simultaneously inputting the multi-sensing monitoring data and the operation state data into the multi-dimensional cyclic neural network model through different input channels, excavating hidden features of different dimensions and outputting predicted residual service life;
the model optimization unit is used for adjusting and optimizing the multidimensional cyclic neural network model by utilizing the prediction deviation between the real residual service life and the output predicted residual service life;
the prediction module comprises:
the data collection unit is used for collecting test data of the large transformer to be predicted, and the test data comprises multi-sensor monitoring data and running state data;
and the prediction unit is used for inputting the test data into the trained multi-dimensional circulation neural network model in real time to obtain the predicted residual service life of the large-scale transformer.
Further, the model construction unit includes:
the historical data collecting subunit is used for collecting historical data of the large transformer and constructing a training data set, wherein the historical data comprises multi-sensor monitoring data, running state data and corresponding real remaining service life;
the data input subunit is used for simultaneously inputting the multi-sensing monitoring data and the running state data in the training data set into the multi-dimensional recurrent neural network model through different input channels;
the characteristic mining subunit is used for mining input data characteristics through a BLSTM layer and a BGRU layer which are parallel to capture hidden characteristics from different dimensions;
and the result output subunit is used for outputting the predicted residual service life of the large transformer from the full connection layer according to the hidden characteristics of the excavated different dimensions.
Further, the data input subunit is configured to perform linear conversion on the input multi-sensor monitoring data in two fully-connected layers to obtain a data hiding feature, and merge the operating state data into the converted multi-sensor monitoring data to construct a high-order vector.
Further, the feature mining subunit employs a dropout technique to prevent repeated capture of the same feature.
Further, the model optimization unit adjusts and optimizes the multidimensional cyclic neural network model by using an adaptive moment estimation algorithm.
In conclusion, the invention provides the method and the system for predicting the residual service life of the transformer under the severe variable working conditions, and the prediction of the residual service life of the large-sized transformer under the severe variable working conditions can be realized. In the multi-dimensional cyclic neural network, the monitoring data and the operation state data of the multi-sensor are fed into the model through different input channels, hidden features from different dimensions are captured, the remaining service life is predicted according to the captured multi-dimensional features, and meanwhile, the model is adjusted and optimized by using the deviation between a real label and a predicted label, so that the accurate prediction of the remaining service life of the large transformer under severe and variable working conditions is realized. The invention simultaneously utilizes the monitoring data and the running state data of the large transformer to carry out modeling analysis, thereby improving the prediction precision under severe and variable working conditions.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the embodiments or the description of the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without inventive labor.
FIG. 1 is a schematic structural diagram of a multidimensional recurrent neural network model provided in an embodiment of the present invention;
fig. 2 is a schematic diagram of a Dropout technique according to an embodiment of the present invention.
Detailed Description
In order to make the objects, features and advantages of the present invention more obvious and understandable, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the embodiments described below are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be obtained by a person skilled in the art without making any creative effort based on the embodiments in the present invention, belong to the protection scope of the present invention.
Aiming at the problem that the prediction performance of the methods on most variable working condition large-scale transformers is poor due to the fact that simultaneous modeling analysis of monitoring data and running state data cannot be achieved in the prior art, the method and the system for predicting the residual service life of the transformer under severe and variable working conditions are provided. In the model training stage, training a multi-dimensional cyclic neural network model by using multi-sensor monitoring data and running state data, excavating hidden features of different dimensions, and outputting predicted residual service life; meanwhile, the multi-dimensional cyclic neural network model is adjusted and optimized by utilizing the prediction deviation between the real residual service life and the output predicted residual service life; and in the prediction stage, inputting the collected test data of the large transformer to be predicted into the trained multidimensional cyclic neural network model in real time to obtain the predicted residual service life of the large transformer. By modeling and analyzing the multi-sensor monitoring data and the operation state data, the residual service life of the large transformer can be accurately predicted.
In order to better understand the technical solutions, the technical solutions will be described in detail below with reference to the drawings and the specific embodiments of the specification, and it should be understood that the embodiments and specific features of the embodiments of the present invention are detailed descriptions of the technical solutions of the present application, and are not limited to the technical solutions of the present application, and the technical features of the embodiments and examples of the present invention may be combined with each other without conflict.
The multidimensional cyclic neural network model disclosed by the invention can execute the prediction task under a single working condition and can complete the prediction task under severe and variable working conditions.
The following describes an embodiment of the method for predicting the remaining service life of the transformer under the severe variable working conditions in detail.
The embodiment provides a method for predicting the residual service life of a transformer under severe variable working conditions, which comprises the following steps: a model training phase and a prediction phase.
In the model training stage, multi-sensor monitoring data and running state data are simultaneously input into a multi-dimensional cyclic neural network model through different input channels, hidden features of different dimensions are mined, and predicted residual service life is output; adjusting and optimizing the multidimensional cyclic neural network model by using the prediction deviation between the real residual service life and the output predicted residual service life;
collecting test data of a large transformer to be predicted in a prediction stage, wherein the test data comprises multi-sensor monitoring data and operation state data; and inputting the test data into the trained multi-dimensional cyclic neural network model in real time to obtain the predicted residual service life of the large transformer.
Specifically, in this embodiment, for the prediction task under severe and variable working conditions, the historical data of the large-scale transformer, including the multi-sensor monitoring data, the operating state data, and the corresponding real remaining service life, is collected first to construct the training set { X } i ,O i ,r i } K And K represents the total number of training samples.X i = { x1, x 2.,. Xt.,. XN } represents a set of input data samples, where x is t ∈R s×1 Is a vector of signal data from s selected sensors at time t. O is i ＝{o 1 ,o 2 ,...,o t ,...,o N Denotes a run-state data sample set, where o t ∈R p×1 Is a vector formed by p running state data at the time t. r is i Denotes AND X at time t i And O i Corresponding true remaining service life.
Specifically, in the present embodiment, the multi-sensor monitoring data and the operating state data in the training data set are simultaneously input into the model through different input channels. The multidimensional recurrent neural network then mines the input data features through the parallel BLSTM and BGRU layers to capture hidden features from different dimensions. And finally, outputting the predicted residual service life of the large transformer from the last full-connection layer. The model structure of the multidimensional recurrent neural network is shown in fig. 1.
Suppose X i ＝{x1,x2,...,xt,...,xN} T Represents the ith input data sample, where x t ∈R s×1 Is a vector of signal data from s selected sensors at time t. O is i ＝{o 1 ,o 2 ,...,o t ,...,o N Denotes a run state data sample set, where o t ∈R p×1 Is a vector formed by p running state data at the time t. As shown in fig. 2, the input multi-sensor monitoring data is first linearly transformed in two fully connected layers to obtain a data hiding representation, which is represented as:
F i ＝W dl X i +b dl (1)
in the formula, W dl And b dl Is the full connection layer weight matrix and the offset. X i A vector of monitoring data representing the input at time i, F i And (3) representing a data hiding representation output by a Dense layer.
Then, the operation state data is spliced to F i In order to construct a high-order vector U i Expressed as follows:
in the formula, O i Representing the input i time running state data vector.
Next, vector U is calculated i Respectively into the stacked BLSTM layer and BGRU layer, and obtain hidden features of different dimensions. The output of the first and middle BLSTM and BGRU layers can be expressed as:
wherein
And &>
Are considered hidden features from different dimensions. Upon obtaining +>
And &>
These hidden feature matrices are then concatenated into a merged feature vector, represented as
Finally, the merged feature vectors are input into two other linear regression dense layers to generate the prediction results.
Wherein,
indicating the output prediction result. W is a group of d2 Representing the weight matrix in the linear regression layer.
It should be noted that, although the method is designed for the multi-condition prediction problem, this does not mean that the method is only suitable for the multi-condition prediction problem. When the single-working-condition prediction problem is processed, the invariance of state data is considered, and the prediction result is not influenced. So that the formula 2 is changed into U i ＝F i To mask off the status data O i The prediction function under the single working condition can be realized.
Further, in this embodiment, after the multidimensional cyclic neural network model is obtained by using the training data set, the obtained multidimensional cyclic neural network model needs to be optimized. And the model is adjusted and optimized by utilizing the prediction deviation between the real residual service life and the predicted residual service life through back propagation and utilizing a reasonable back propagation strategy and an optimization algorithm. Considering that overfitting of the model is caused by too small data sample size or too complex model structure, a corresponding model regularization method needs to be designed.
The optimization algorithm of the model parameters can directly influence the efficiency of the multidimensional cyclic neural network training. Therefore, an Adaptive moment estimation (Adam) algorithm is adopted to replace a traditional Stochastic Gradient Descent (SGD) optimizer so as to minimize the loss function of the multidimensional recurrent neural network. In the model parameter optimization process, the traditional SGD algorithm maintains a fixed learning rate to update all parameters, which results in extremely inefficient update of network parameters. The network parameter updating process of the Adam algorithm is represented as follows:
m t ＝β 1 m t-1 +(1-β 1 )g t (10)
wherein, g t Is the gradient of the loss function L (theta) over the network parameter set theta. m is t And n t First order moment estimates and second order moment estimates of the gradient, respectively.
And &>
Are each m t And n t The result of the offset correction of (1).
And &>
The exponential decay rates of the first moment estimate and the second moment estimate, respectively. η represents the step length and e represents the numerical stability constant. Theta t Is the calculated theta t-1 The updated value at time t.
Further, increasing the number of layers of the neural network increases the model training time and increases the risk of overfitting. Overfitting results in a neural network that performs well on the training dataset and poorly on the test dataset. To overcome this problem, the dropout technique is employed in the multidimensional recurrent neural network of the present invention to prevent the same feature from being repeatedly captured. A schematic diagram of the dropout technique is shown in FIG. 2. In this schematic, the darker circles are constructed during the training of the multidimensional recurrent neural network to obtain data from the monitored signals to construct the high-dimensional input vectors. Once the time window completes the acquisition, it is slid in time by one sample period to perform the next acquisition until the signal ends.
Further, in the model training phase, input data is input into the multidimensional cyclic neural network model, and the model is forwarded according to the processes of formulas 1 to 9. In order to realize the back propagation of the multidimensional recurrent neural network, an appropriate model loss function needs to be designed for the multidimensional recurrent neural network model. Prediction value based on model output
And the true data value r i Introducing a mean square error function, and designing a loss function as follows:
wherein, F MDRNN (X i ，O i And theta) represents that the monitor data X is to be sensed i And operating state data O i And inputting the prediction output obtained by the multidimensional cyclic neural network model, wherein theta represents a multidimensional cyclic neural network model parameter set.
Further, in the embodiment, in the testing process, online test data is preprocessed and then input into the trained multidimensional cyclic neural network model in real time, so that the predicted remaining service life of the large transformer can be obtained. Wherein, the input data mainly is transformer monitoring data, includes: monitoring the temperature and load of transformer oil, signals of dissolved gas and micro water in the oil, grounding current of an iron core, vibration signals of a transformer and the like; the output data is the remaining service life of the transformer.
Correspondingly, the invention also provides a prediction system for the residual service life of the transformer under severe and variable working conditions, which comprises a model training module and a prediction module.
Specifically, in this embodiment, the model training module includes:
the model building unit is used for simultaneously inputting the multi-sensing monitoring data and the running state data into the multi-dimensional recurrent neural network model through different input channels, mining hidden features of different dimensions and outputting predicted residual service life;
and the model optimization unit is used for adjusting and optimizing the multi-dimensional recurrent neural network model by utilizing the prediction deviation between the real residual service life and the output predicted residual service life, and adjusting and optimizing the multi-dimensional recurrent neural network model by using an adaptive moment estimation algorithm.
Specifically, in this embodiment, the prediction module includes:
the system comprises a data collection unit, a prediction unit and a prediction unit, wherein the data collection unit is used for collecting test data of the large transformer to be predicted, and the test data comprises multi-sensor monitoring data and running state data;
and the prediction unit is used for inputting the test data into the trained multidimensional cyclic neural network model in real time to obtain the predicted residual service life of the large transformer.
Further, in this embodiment, the model construction unit includes:
the historical data collecting subunit is used for collecting historical data of the large transformer and constructing a training data set, wherein the historical data comprises multi-sensor monitoring data, running state data and corresponding real remaining service life;
the data input subunit is used for simultaneously inputting the multi-sensing monitoring data and the running state data in the training data set into the multi-dimensional cyclic neural network model through different input channels, wherein the input multi-sensing monitoring data is firstly subjected to linear conversion in two fully-connected layers to obtain data hidden features, and then the running state data is merged into the converted multi-sensing monitoring data to construct a high-order vector;
the characteristic mining subunit is used for mining input data characteristics through a BLSTM layer and a BGRU layer which are parallel to capture hidden characteristics from different dimensions, and a dropout technology is adopted to prevent the same characteristics from being repeatedly captured;
and the result output subunit is used for outputting the predicted residual service life of the large-scale transformer from the full connection layer according to the hidden characteristics of different dimensionalities of excavation.
The above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (10)
1. The method for predicting the residual service life of the transformer under severe and variable working conditions is characterized by comprising the following steps of: a model training phase and a prediction phase;
the model training phase comprises:
s11, simultaneously inputting multi-sensor monitoring data and running state data into a multi-dimensional recurrent neural network model through different input channels, mining hidden features of different dimensions, and outputting predicted residual service life;
s12, adjusting and optimizing the multidimensional cyclic neural network model by using the prediction deviation between the real residual service life and the output predicted residual service life;
the prediction phase comprises:
s21, collecting test data of the large transformer to be predicted, wherein the test data comprises multi-sensor monitoring data and running state data;
and S22, inputting the test data into the trained multidimensional cyclic neural network model in real time to obtain the predicted residual service life of the large transformer.
2. The method for predicting the residual service life of the transformer under the severe variable working conditions according to claim 1, wherein the step S11 comprises the following steps:
s111, collecting historical data of the large transformer, and constructing a training data set, wherein the historical data comprises multi-sensor monitoring data, running state data and corresponding real remaining service life;
step S112, simultaneously inputting the multi-sensing monitoring data and the running state data in the training data set into the multi-dimensional recurrent neural network model through different input channels;
step S113, mining input data characteristics through a BLSTM layer and a BGRU layer which are parallel to each other so as to capture hidden characteristics from different dimensions;
and S114, outputting the predicted residual service life of the large transformer from the full connection layer according to the hidden features of different dimensions of excavation.
3. The method for predicting the remaining service life of the transformer under the severe variable working conditions according to claim 2, wherein the step S112 comprises:
step S1121, firstly performing linear conversion on the input multi-sensor monitoring data in two full-connection layers to obtain data hiding characteristics;
and step S1122, merging the running state data into the converted multi-sensing monitoring data to construct a high-order vector.
4. The method for predicting the remaining service life of the transformer under the severe variable working conditions according to claim 2, wherein in step S113, a dropout technique is adopted to prevent the same characteristics from being repeatedly captured.
5. The method for predicting the remaining service life of the transformer under the severe and variable working conditions according to claim 1, wherein in step S12, a multi-dimensional recurrent neural network model is adjusted and optimized by using an adaptive moment estimation algorithm.
6. A transformer remains life prediction system under being used for abominable changeable operating mode, its characterized in that includes: a model training module and a prediction module;
the model training module comprises:
the model construction unit is used for simultaneously inputting the multi-sensing monitoring data and the operation state data into the multi-dimensional cyclic neural network model through different input channels, excavating hidden features of different dimensions and outputting predicted residual service life;
the model optimization unit is used for adjusting and optimizing the multidimensional cyclic neural network model by utilizing the prediction deviation between the real residual service life and the output predicted residual service life;
the prediction module comprises:
the system comprises a data collection unit, a prediction unit and a prediction unit, wherein the data collection unit is used for collecting test data of the large transformer to be predicted, and the test data comprises multi-sensor monitoring data and running state data;
and the prediction unit is used for inputting the test data into the trained multidimensional cyclic neural network model in real time to obtain the predicted residual service life of the large transformer.
7. The system for predicting the residual service life of the transformer under the severe variable working conditions according to claim 6, wherein the model building unit comprises:
the historical data collecting subunit is used for collecting historical data of the large transformer and constructing a training data set, wherein the historical data comprises multi-sensor monitoring data, running state data and corresponding real remaining service life;
the data input subunit is used for simultaneously inputting the multi-sensing monitoring data and the running state data in the training data set into the multi-dimensional recurrent neural network model through different input channels;
the characteristic mining subunit is used for mining input data characteristics through a BLSTM layer and a BGRU layer which are parallel to capture hidden characteristics from different dimensions;
and the result output subunit is used for outputting the predicted residual service life of the large transformer from the full connection layer according to the hidden characteristics of the excavated different dimensions.
8. The system for predicting the residual service life of the transformer under the severe and variable working conditions according to claim 7, wherein the data input subunit is configured to perform linear conversion on input multi-sensor monitoring data in two fully-connected layers to obtain data hiding characteristics, and merge operating state data into the converted multi-sensor monitoring data to construct a high-order vector.
9. The system for predicting the remaining service life of the transformer under the severe variable working conditions according to claim 7, wherein the feature mining subunit adopts a dropout technology to prevent the same feature from being captured repeatedly.
10. The system for predicting the remaining service life of the transformer under the severe variable working conditions according to claim 6, wherein the model optimization unit adjusts and optimizes the multidimensional cyclic neural network model by using an adaptive moment estimation algorithm.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211531192.XA CN115906648A (en) | 2022-12-01 | 2022-12-01 | Method and system for predicting remaining service life of transformer under severe variable working conditions |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211531192.XA CN115906648A (en) | 2022-12-01 | 2022-12-01 | Method and system for predicting remaining service life of transformer under severe variable working conditions |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115906648A true CN115906648A (en) | 2023-04-04 |
Family
ID=86495255
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211531192.XA Pending CN115906648A (en) | 2022-12-01 | 2022-12-01 | Method and system for predicting remaining service life of transformer under severe variable working conditions |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115906648A (en) |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111460728A (en) * | 2020-03-09 | 2020-07-28 | 华南理工大学 | Method and device for predicting residual life of industrial equipment, storage medium and equipment |
-
2022
- 2022-12-01 CN CN202211531192.XA patent/CN115906648A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111460728A (en) * | 2020-03-09 | 2020-07-28 | 华南理工大学 | Method and device for predicting residual life of industrial equipment, storage medium and equipment |
Non-Patent Citations (1)
| Title |
|---|
| YIWEI CHENG 等: "Multi-dimensional recurrent neural network for remaining useful life prediction under variable operating conditions and multiple fault modes", 《APPLIED SOFT COMPUTING》, vol. 118, pages 1 - 12 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111596604B (en) | Intelligent fault diagnosis and self-healing control system and method for engineering equipment based on digital twinning | |
| CN109934337B (en) | Spacecraft telemetry data anomaly detection method based on integrated LSTM | |
| CN114169545B (en) | Method, device, medium and equipment for intelligent fault diagnosis and operation and maintenance guide of thermal power plant equipment | |
| JP2019527413A (en) | Computer system and method for performing root cause analysis to build a predictive model of rare event occurrences in plant-wide operations | |
| CN110210169B (en) | LSTM-based shield tunneling machine fault prediction method | |
| CN113378939B (en) | Structure digital twin modeling and parameter identification method based on physical driving neural network | |
| CN102361014A (en) | State monitoring and fault diagnosis method for large-scale semiconductor manufacture process | |
| CN113325721B (en) | Model-free adaptive control method and system for industrial system | |
| CN113094860B (en) | Industrial control network flow modeling method based on attention mechanism | |
| CN113723010A (en) | Bridge damage early warning method based on LSTM temperature-displacement correlation model | |
| CN111967185A (en) | Cable-stayed bridge state evaluation method based on cable force and displacement distribution correlation modeling | |
| CN113987908A (en) | Natural gas pipe network leakage early warning method based on machine learning method | |
| CN113759868A (en) | Industrial Ethernet fault prediction method based on neural network | |
| CN116591768A (en) | Tunnel monitoring method, system and device based on distributed network | |
| CN116614177A (en) | Optical fiber state multidimensional parameter monitoring system | |
| CN115659249B (en) | Abnormality detection method for intelligent station measurement control system | |
| CN110750455B (en) | Intelligent online self-updating fault diagnosis method and system based on system log analysis | |
| CN112763215A (en) | Multi-working-condition online fault diagnosis method based on modular federal deep learning | |
| CN114880917A (en) | Method and device for building health state model and predicting performance trend of pumped storage unit | |
| CN111177916A (en) | Satellite electromechanical product service life prediction method based on machine deep learning | |
| CN115906648A (en) | Method and system for predicting remaining service life of transformer under severe variable working conditions | |
| CN114819350B (en) | RUL prediction method based on multitasking sensor fault condition | |
| CN117977536A (en) | Smart power grid load identification method | |
| CN111524348A (en) | Long-short term traffic flow prediction model and method | |
| CN114880767B (en) | Aero-engine residual service life prediction method based on attention mechanism Dense-GRU network |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20230404 |
|
| RJ01 | Rejection of invention patent application after publication |