CN-122024437-A - Transformer risk real-time early warning method based on improved transformer algorithm

CN122024437ACN 122024437 ACN122024437 ACN 122024437ACN-122024437-A

Abstract

The invention relates to a transformer risk real-time early warning method based on an improved transformer algorithm, which comprises the steps of firstly realizing space-time alignment fusion of data sets, then constructing a transformer state evaluation index system based on a multistage fuzzy comprehensive research and judgment method, weighting the transformer state data sets to obtain optimized transformer data sets so as to obtain the safety threshold value of each transformer state parameter, meanwhile, demarcating the early warning risk level of the transformer operation parameters by the method, improving the data richness, finally carrying out parameter optimization on the transformer algorithm by TAAO, completing the prediction of each state parameter of the transformer, combining with the demarcation of the early warning risk level, and realizing the real-time judgment of the active early warning of the transformer risk.

Inventors

MA JIN
XU MING
HE WENPING
WANG WENBIN
ZHANG XIAOHUI
FENG ZHENG
NING JINFENG

Assignees

国网山西省电力公司长治供电公司

Dates

Publication Date: 20260512
Application Date: 20251231

Claims (8)

1. A transformer risk real-time early warning method based on an improved transformer algorithm is characterized by comprising the following steps of: Step 1, collecting transformer data and preprocessing the data; Step 2, carrying out multistage fuzzy evaluation on the transformer data preprocessed in the step 1, and constructing a transformer state evaluation index system; Step 3, constructing an MFCE-TAAO-transducer algorithm model according to the data preprocessed in the step 1 and the transformer state evaluation index system constructed in the step 2, and training; And 4, performing real-time early warning on the risk of the transformer by using the trained MFCE-TAAO-transducer algorithm model.
2. The transformer risk real-time early warning method based on the improved transformer algorithm of claim 1, wherein the step 1 comprises the following steps: Step 1.1, collecting transformer oil chromatographic data, oil temperature data, winding temperature data, three-phase current and three-phase voltage data; and step 1.2, carrying out space-time alignment fusion and dynamic threshold balance on the data acquired in the step 1.1.
3. The transformer risk real-time early warning method based on the improved transformer algorithm of claim 2, wherein the step 1.2 comprises the following steps: The method comprises the steps of 1.2.1, fusing the acquired transformer data in the step 1.1 to form multi-mode vector representation, fusing the acquired data, operation records and maintenance logs into a unified feature space by adopting a feature fusion method, converting all time stamps into YYYY-MM-DD HH: MM: SS format, and simultaneously aligning transformer oil chromatographic data, oil temperature, winding temperature, three-phase voltage and three-phase current data time stamps closest to transformer oil chromatographic test data time stamps by adopting a nearest neighbor matching algorithm according to the condition of different data frequencies to finish the time-space alignment fusion of the transformer data; step 1.2.2, dynamic threshold balance is carried out on the space-time alignment fusion data obtained in the step 1.2.1 by using a moving average algorithm, and the data are smoothed and trends are identified; And step 1.2.3, dividing the dynamic threshold balanced data into a training set and a testing set.
4. The transformer risk real-time early warning method based on the improved transformer algorithm of claim 1, wherein the step 2 comprises the following steps: Step 2.1, determining a transformer state evaluation index system based on the transformer data preprocessed in the step 1, and taking transformer oil color spectrum, temperature, three-phase voltage and three-phase current data as an evaluation index set U= { oil color spectrum, oil temperature, winding temperature, three-phase voltage and three-phase current }, in the transformer state index system; step 2.2, determining each evaluation index weight A by adopting an entropy weight method; Step 2.3, constructing a membership function to obtain a first-level fuzzy vector, constructing a multi-level fuzzy matrix B, and determining the score of an evaluation target according to the first-level fuzzy vector; And 2.4, based on the score of the evaluation target obtained in the step 2.3, performing simulation evaluation on the health degree of the transformer by adopting a comprehensive research and judgment method aiming at transformer oil chromatography, temperature and voltage, thereby obtaining a health degree evaluation standard of the transformer.
5. The transformer risk real-time early warning method based on the improved transformer algorithm of claim 4, wherein the step 2.3 comprises the following steps: Step 2.3.1, constructing a fuzzy evaluation matrix, wherein the fuzzy evaluation matrix is formed by a membership subset R= { R 1 ,r 2 ,…,r m } of each index of the transformer, and a semi-trapezoidal distribution function is selected as a membership function R m : ; ; Wherein m is the number of state indexes of the transformer, V is the number of comment sets which form a comment set number standard V= {0,25,50, 75, 100}, and x is the number of data sets of the transformer, namely specific numbers of oil chromatogram, oil temperature, winding temperature, three-phase voltage and three-phase current; Step 2.3.2, obtaining a first-order fuzzy vector B i according to a membership function r m : ; Wherein R ij represents the membership degree of the ith transformer evaluation index to the jth comment set value, R is a multidimensional fuzzy relation matrix, n is the number of the comment sets, and m is the number of the transformer state indexes; ; Wherein A is an entropy weight method for determining each index weight, and b m represents the membership degree of the transformer data to the m-th comment set value; step 2.3.3, obtaining an evaluation target score T according to the first-order fuzzy vector B i : ; Wherein m is the number of the state indexes of the transformer, V is the value of the comment set, which forms a comment set value standard V= {0,25,50, 75, 100}, and b j represents the membership degree of the transformer data to the value of the j-th comment set.
6. The transformer risk real-time early warning method based on the improved transformer algorithm of claim 1, wherein the step 3 comprises the following steps: Step 3.1, converting the preprocessed data in the step 1 into multi-mode vectors, and inputting a transducer model; step 3.2, optimizing parameters of the transducer model through TAAO to obtain optimal parameters; And 3.3, inputting the training set into the TAAO-transducer model optimized in the step 3.2 to obtain transformer prediction data, and simultaneously comparing the transformer prediction data with original test set data to obtain Root Mean Square Error (RMSE), average percentage error (MAPE) and a decision coefficient R 2 , and obtaining a trained MFCE-TAAO-transducer algorithm model.
7. The transformer risk real-time early warning method based on the improved transformer algorithm of claim 6, wherein the specific implementation method of the step 3.2 is that the transformer algorithm is subjected to gradient decomposition of a loss function from three angles of trend, periodicity and noise characteristics of transformer data, and the gradient of the loss function L (theta) Decomposition by sliding window fourier transform: ; The trend component g trend corresponds to a low-frequency part and is used for reflecting the increasing and decreasing trend of the transformer data, the cyclic component g cycle corresponds to the inverse transformation of the significant frequency and is used for reflecting the periodicity of the transformer data, the noise component g noise corresponds to the residual high-frequency part and is used for reflecting the influence of an emergency on the transformer data, k is the number of detected dominant periods, and t is the number of cyclic steps; the specific calculation method of the trend component of the transformer comprises the following steps: ; where a is the slope parameter of the transformer data trend fit, τ is the time step index, ω is the sliding window size, g τ is the gradient of the loss function at time tau, the L2 norm is represented by the L; The specific calculation formula of the circulating component of the transformer is as follows: ; Wherein F is Fourier transform operation, T k is time step, and F is frequency; The noise component of the transformer is a residual high-frequency part corresponding to the loss function and is used for reflecting the influence of sudden events on the state prediction of the transformer, k is the detected number of leading periods, and t is the number of circulating steps; the three components of the transformer are weighted and updated: ; Wherein θ t is a model parameter of the t-th step, η t represents a global learning rate, α t is a trend weight, β k,t is a weight of the kth periodic component, γ t is a noise weight, sign () is a sign function for controlling an update direction of noise; The weight coefficient calculation method of different characteristics comprises the following steps: Trend weight: ; Wherein g t is the gradient of the loss function L (θ) Cyclic weight: ; Wherein σ 2 represents the variance of the periodic component; Noise weight: ; Wherein the denominator is used to ensure attenuation; decoupling constraint is realized on trend components and cyclic components of the transformation state vector, and pattern confusion is avoided by adding regularization terms: ; Wherein lambda is regularization coefficient, the value of lambda is 0.01, W i trend is trend component of the ith layer parameter, and W i cycle is periodic component of the ith layer parameter.
8. The transformer risk real-time early warning method based on the improved transformer algorithm of claim 6, wherein the method for calculating the Root Mean Square Error (RMSE) in the step 3.3 is as follows: ; ; ; Wherein N is the total data amount, Y i is the predicted value of the ith transformer data, and Y i is the true value of the ith transformer data; The range of the three model evaluation indexes all belong to [0, ++ infinity), and the three model evaluation indexes of MAPE and R 2 belong to [0,1], the smaller the model evaluation indexes of RMSE, MAPE and R 2 are, the better the model prediction effect is, the higher the accuracy is, and on the contrary, the larger the model prediction is, the worse the model prediction is, the lower the accuracy is.

Description

Transformer risk real-time early warning method based on improved transformer algorithm Technical Field The invention belongs to the technical field of transformer operation reliability monitoring, and particularly relates to a transformer risk real-time early warning method based on an improved transformer algorithm. Background Transformers serve as core devices in power systems and take on important roles in power transformation and transmission. With the continuous expansion of the power grid scale and the increasing demand for power, the operational reliability of transformers has a crucial impact on the safety and stability of the power system. However, as the service life of the transformer increases, the failure rate thereof also shows a gradual rise. Therefore, the state evaluation is carried out on the transformer, potential faults are found and processed in time, and the method becomes a key link for ensuring the safe operation of the power system. The transformer fault types are various, and mainly comprise insulation faults, winding faults, iron core faults, tap switch faults and the like. These faults often occur in connection with various links of design, manufacture, installation, operation and maintenance of the transformer. Insulation faults are one of the most common types of faults of transformers, and are mainly caused by degradation of insulation performance due to aging, moisture, pollution or mechanical damage of insulation materials. The winding failure may be caused by poor winding manufacturing process, winding deformation, short-circuit current surge, and the like. Core failure may be related to core material quality, core clamping force, magnetic circuit design, etc. The tap switch failure may be caused by poor contact, mechanical jamming, contact wear, etc. At present, the intelligent fault prediction method for the transformer in the existing research is mainly concentrated in the field of classifiers related to oil chromatographic data, and the real-time risk early warning method for the state of the transformer is still in a development stage, and the specific problems mainly include the following three points: 1. At present, fault detection data of a transformer are limited to oil chromatographic data, the richness of data sources is insufficient, the data diversity is low (limited to gas concentration), and a good effect cannot be achieved on real-time early warning of the state of the transformer. 2. The optimization method of the time sequence prediction algorithm for real-time early warning of the state risk of the transformer in the existing research mostly has the problem of blind average, and cannot be effectively applied to industrial environments such as the state time sequence prediction of the transformer. 3. The existing transformer model has low prediction precision on transformer data, and has the phenomenon of false alarm and missing alarm. Disclosure of Invention The invention aims to overcome the defects of the prior art, provides a transformer risk real-time early warning method based on an improved transformer algorithm, and can solve the problems of low data diversity, low transformer real-time risk early warning accuracy and the like in the existing transformer fault research. The invention solves the technical problems by adopting the following technical scheme: A transformer risk real-time early warning method based on an improved transformer algorithm comprises the following steps: Step 1, collecting transformer data and preprocessing the data; Step 2, carrying out multistage fuzzy evaluation on the transformer data preprocessed in the step 1, and constructing a transformer state evaluation index system; Step 3, constructing an MFCE-TAAO-transducer algorithm model according to the data preprocessed in the step 1 and the transformer state evaluation index system constructed in the step 2, and training; And 4, performing real-time early warning on the risk of the transformer by using the trained MFCE-TAAO-transducer algorithm model. Furthermore, the step1 includes the steps of: Step 1.1, collecting transformer oil chromatographic data, oil temperature data, winding temperature data, three-phase current and three-phase voltage data; and step 1.2, carrying out space-time alignment fusion and dynamic threshold balance on the data acquired in the step 1.1. Furthermore, said step 1.2 comprises the steps of: The method comprises the steps of 1.2.1, fusing the acquired transformer data in the step 1.1 to form multi-mode vector representation, fusing the acquired data, operation records, maintenance logs and other data into a unified feature space by adopting a feature fusion method, converting all time stamps into YYYY-MM-DD HH: MM: SS format, and simultaneously aligning transformer oil chromatographic data, oil temperature, winding temperature, three-phase voltage and three-phase current data time stamps closest to transformer oil chromatographic test data ti