CN-121998614-A - Full life cycle management method of oil immersed transformer based on big data analysis

CN121998614ACN 121998614 ACN121998614 ACN 121998614ACN-121998614-A

Abstract

The application relates to the technical field of power equipment state evaluation, and particularly discloses a full life cycle management method of an oil immersed transformer based on big data analysis, which comprises the steps of obtaining real-time operation data of a current transformer, constructing a real-time input vector and a real-time output vector, wherein each component of the real-time input vector can influence the corresponding component of the real-time output vector; and calculating the health state score of the current transformer according to the real-time state matrix, the health state matrix and the retired state matrix, and formulating a maintenance strategy according to the health state score. The method provided by the application adopts a fuzzy comprehensive evaluation mode of a real-time state matrix, so that the corresponding relation between the input and the output of the transformer is reserved, and the interference of a single factor on the health state evaluation is avoided, thereby realizing accurate and low-cost real-time health state evaluation.

Inventors

JIAO YUXIAN
LI YE
Hu Yahao
MA SHUO
HOU CHUANG
Duan Yunzhe
Si Guangpo
WANG ZHENHUA
XIAO BING
XIE CHUNJING
WANG LIN
XIAO SHULEI
DUAN LIAN
CHEN KUI
TIAN XIAOJING

Assignees

河南天力电气设备有限公司

Dates

Publication Date: 20260508
Application Date: 20260116

Claims (9)

1. The full life cycle management method of the oil immersed transformer based on big data analysis is characterized by comprising the following steps of: Acquiring real-time operation data of a current transformer, and constructing a real-time input vector and a real-time output vector, wherein each component of the real-time input vector can influence the corresponding component of the real-time output vector; estimating a real-time state matrix through a state estimation model according to the real-time input vector and the real-time output vector; And calculating the health state score of the current transformer according to the real-time state matrix, the health state matrix and the retired state matrix, and formulating a maintenance strategy according to the health state score.
2. The method of claim 1, wherein each component of the real-time input vector represents a voltage and a current, respectively, on a high voltage side and each component of the real-time output vector represents a voltage and a current, respectively, on a low voltage side.
3. The method of claim 1, wherein the state estimation model estimates the real-time state matrix by: establishing an ideal mathematical model of the current transformer; Calculating a theoretical output vector in an ideal state according to the real-time input vector by utilizing the ideal mathematical model; calculating a real-time error vector according to the theoretical output vector and the real-time output vector; and calculating a real-time state matrix according to the real-time input vector, the real-time output vector and the real-time error vector.
4. The method of claim 1, wherein the retirement state matrix is obtained by: respectively aiming at different transformers, establishing a historical health state matrix and a historical retired state matrix according to historical operation data of the transformers; Dividing a plurality of different transformer history health state matrixes and history retired state matrixes into a training set and a testing set, training a machine learning model through the training set, and further fine-tuning super-parameters of the machine learning model through the testing set; And deducing the retired state matrix of the current transformer according to the health state matrix of the current transformer by using the trained machine learning model.
5. The method of claim 1, wherein the health status score is calculated by: Differencing the real-time state matrix and the health state matrix to obtain a health difference matrix, and calculating the F-norm of the health difference matrix; Differencing the real-time state matrix and the retired state matrix to obtain a retired difference matrix, and calculating the F-norm of the retired difference matrix; Dividing the F-norm of the health difference matrix by the sum of the F-norm of the health difference matrix and the F-norm of the retired difference matrix to obtain the health state score of the current transformer.
6. The method of claim 1, wherein the maintenance policy comprises: When the health status score is within a normal threshold interval, no maintenance is required; When the health status score is within a good threshold interval, performing regular maintenance according to a daily maintenance plan; when the health state score is within a general threshold interval, maintenance is needed; when the health state score is within a critical threshold interval, the maintenance needs to be carried out as soon as possible; When the health status score is within a fault threshold interval, immediate service is required.
7. The method of claim 1, wherein when the health status score is below a fault threshold, performing the steps of: Acquiring real-time operation data of the current transformer after the health state score is lower than a fault threshold value; the deep learning model derives a real-time relevance matrix according to the real-time operation data; matching the real-time association degree matrix with standard association degree matrices of different fault types, if the matching is successful, sending out a corresponding fault type prompt, otherwise, giving out the similarity between the real-time association degree matrix and the standard association degree matrix of at least one fault type.
8. The method of claim 7, wherein the standard association matrix is obtained by: Classifying the historical operation data according to different fault types according to the actual fault records of the transformer; Respectively aiming at the historical operation data of each fault type, grading the health state, and intercepting the historical operation data from when the health state grading is smaller than a fault threshold value to when the fault occurs; and analyzing according to the intercepted historical operation data by using a deep learning algorithm to obtain the standard association degree matrix of the fault type.
9. The method of claim 7, wherein determining whether the real-time relevance matrix matches the standard relevance matrix is performed by: The real-time correlation matrix is subjected to difference with the standard correlation matrix to obtain a similarity matrix, and the F-norm of the similarity matrix is calculated; f-norms of the standard association degree matrix are calculated; and calculating the ratio of the F-norm of the similarity matrix to the F-norm of the standard association matrix, and if the ratio is smaller than a matching threshold, judging that the real-time association matrix is matched with the standard association matrix.

Description

Full life cycle management method of oil immersed transformer based on big data analysis Technical Field The invention relates to the technical field of power equipment state evaluation, in particular to a full life cycle management method of an oil immersed transformer based on big data analysis. Background The full life cycle management of the transformer mainly comprises four stages of design and manufacture, installation and debugging, operation and maintenance and retirement treatment, wherein the operation and maintenance stage is the longest stage with highest cost in the life cycle and is the core of modern management, so that the establishment of a reasonable overhaul and maintenance strategy is an effective means for reducing the cost of the operation and maintenance stage. In the daily use process of the transformer, the maintenance strategy of the transformer is mostly realized through regular manual inspection, and maintenance personnel judge the operation state of the transformer by observing a plurality of operation indexes (such as oil level, oil temperature, noise and the like) or appearance of the transformer. The potential fault hidden trouble existing in the transformer cannot be found effectively and timely by means of manual inspection, and the problems of low efficiency, high subjectivity and the like of manual inspection are solved, so that the maintenance requirement of a large-scale transformer group is difficult to meet. At present, some prior art attempts have been made to monitor a transformer in real time by using a big data analysis method, for example, the prior art CN 109711663B discloses a method and a system for evaluating and correcting the state of an oil immersed transformer of a transformer substation based on big data analysis, the method realizes comprehensive evaluation of the state of the transformer through steps of multi-source information fusion, principal component analysis dimension reduction, evidence theory fusion, iterative correction and the like, however, some state quantities in the scheme provided by the prior art need to be measured through chemical experiments or high-pressure experiments, which definitely increases the state monitoring cost of the transformer. And the transformer state parameter multidimensional criterion matrix is subjected to dimension reduction processing by a principal component analysis method, so that a transformer state evaluation index is established, the efficiency of data processing is improved, the relation between data is linearized in the dimension reduction process, important information loss can be possibly caused, and the robustness of the result is reduced. Disclosure of Invention In order to solve the technical problems of high state evaluation cost and easy loss of important information of a transformer in the prior art, the application provides a full life cycle management method of an oil immersed transformer based on big data analysis, which comprises the following steps: Acquiring real-time operation data of a current transformer, and constructing a real-time input vector and a real-time output vector, wherein each component of the real-time input vector can influence the corresponding component of the real-time output vector; estimating a real-time state matrix through a state estimation model according to the real-time input vector and the real-time output vector; And calculating the health state score of the current transformer according to the real-time state matrix, the health state matrix and the retired state matrix, and formulating a maintenance strategy according to the health state score. Specifically, each component of the real-time input vector represents a voltage and a current on the high-voltage side, respectively, and each component of the real-time output vector represents a voltage and a current on the low-voltage side, respectively. Specifically, the state estimation model estimates a real-time state matrix by: establishing an ideal mathematical model of the current transformer; Calculating a theoretical output vector in an ideal state according to the real-time input vector by utilizing the ideal mathematical model; calculating a real-time error vector according to the theoretical output vector and the real-time output vector; and calculating a real-time state matrix according to the real-time input vector, the real-time output vector and the real-time error vector. Specifically, the retirement state matrix is obtained by the following steps: respectively aiming at different transformers, establishing a historical health state matrix and a historical retired state matrix according to historical operation data of the transformers; Dividing a plurality of different transformer history health state matrixes and history retired state matrixes into a training set and a testing set, training a machine learning model through the training set, and further fine-tuning super-parameters of the machine