CN108390380B - Method and system for predicting state parameter trend of transformer - Google Patents

Method and system for predicting state parameter trend of transformer Download PDF

Info

Publication number
CN108390380B
CN108390380B CN201810468349.6A CN201810468349A CN108390380B CN 108390380 B CN108390380 B CN 108390380B CN 201810468349 A CN201810468349 A CN 201810468349A CN 108390380 B CN108390380 B CN 108390380B
Authority
CN
China
Prior art keywords
transformer
monitoring state
state quantity
online monitoring
prediction
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.)
Active
Application number
CN201810468349.6A
Other languages
Chinese (zh)
Other versions
CN108390380A (en
Inventor
盛戈皞
代杰杰
李鹏
吴绍军
李金忠
李文升
安树怀
张书琦
王健一
高飞
仇宇舟
汪可
侯慧娟
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shanghai Jiaotong University
China Electric Power Research Institute Co Ltd CEPRI
Qingdao Power Supply Co of State Grid Shandong Electric Power Co Ltd
Original Assignee
Shanghai Jiaotong University
China Electric Power Research Institute Co Ltd CEPRI
Qingdao Power Supply Co of State Grid Shandong Electric Power Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Shanghai Jiaotong University, China Electric Power Research Institute Co Ltd CEPRI, Qingdao Power Supply Co of State Grid Shandong Electric Power Co Ltd filed Critical Shanghai Jiaotong University
Priority to CN201810468349.6A priority Critical patent/CN108390380B/en
Publication of CN108390380A publication Critical patent/CN108390380A/en
Application granted granted Critical
Publication of CN108390380B publication Critical patent/CN108390380B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J3/00—Circuit arrangements for ac mains or ac distribution networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J2203/00—Indexing scheme relating to details of circuit arrangements for AC mains or AC distribution networks
    • H02J2203/20—Simulating, e g planning, reliability check, modelling or computer assisted design [CAD]
    • 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
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
    • Y02E60/00—Enabling technologies; Technologies with a potential or indirect contribution to GHG emissions mitigation

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Power Engineering (AREA)
  • Computer Hardware Design (AREA)
  • Evolutionary Computation (AREA)
  • Geometry (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a transformer state parameter trend prediction method, which comprises the following steps: (1) collecting the on-line monitoring state quantity of the transformer; (2) carrying out normalization processing on the transformer online monitoring state quantity by adopting a dispersion normalization method to obtain a transformer online monitoring state quantity matrix X; (3) constructing a prediction model and determining the structure of the prediction model, wherein the prediction model comprises a feature extractor and a feedforward neural network layer (4), and the prediction model is trained by taking part of transformer online monitoring state quantity in a transformer online monitoring state quantity matrix X as input; (5) inputting the other part of the transformer online monitoring state quantity in the transformer online monitoring state quantity matrix X as input data into a trained prediction model, and inputting the characteristic parameters between the input data and a prediction result into a feedforward neural network layer after extracting characteristic parameters by a characteristic extractor; (6) and outputting a prediction result. In addition, the invention also discloses a transformer state parameter trend prediction system.

Description

Method and system for predicting state parameter trend of transformer
Technical Field
The present invention relates to a prediction method and system, and more particularly, to a prediction method and system for trend prediction.
Background
The transformer state is generally represented by the change of a plurality of characteristic parameters, the characteristic parameters (namely transformer state parameters) can correspondingly change along with different factors of faults, such as fault types, fault occurrence positions and fault severity, and the change trend of key characteristic parameters corresponding to different faults is predicted, so that the change process of the transformer state can be clearly and intuitively known.
Because the data of the transformer state parameters can be regarded as a complex data sequence, at present, the conventional prediction methods usually use a single or few statistical formulas (such as regression analysis or time sequence analysis) of the parameters to extrapolate the future time sequence and the change trend, but these prediction methods cannot consider the influence of a plurality of relevant factors on the formula adaptability, that is, once the data of a certain transformer state parameter is missing, prediction cannot be performed.
With the development of the smart grid technology, more and more information such as transformer state data, grid operation data and meteorological information data are gradually integrated and shared on a unified information platform, so that a prediction method is expected to be obtained, wherein the correlation relation mining can be performed on multi-source information related to the transformer state, the correlation relation between state quantities obtained by mining is used as priori knowledge, the prediction model is adaptively adjusted to improve the prediction precision of the transformer state parameter trend, and the effect of accurately predicting the transformer state parameter trend is finally achieved.
Disclosure of Invention
The invention aims to provide a transformer state parameter trend prediction method, which extracts the incidence relation between on-line monitoring state quantities of each transformer by using a grid long-and-short time memory network, extracts the trend relation between the incidence relation and the future state quantity (namely the required prediction result), trains a prediction model, and finally outputs the prediction result through a feedforward neural network layer in the trained prediction model so as to accurately predict the transformer state parameter trend.
Based on the above purpose, the invention provides a transformer state parameter trend prediction method, which comprises the following steps:
(1) collecting the on-line monitoring state quantity of the transformer;
(2) carrying out normalization processing on the transformer online monitoring state quantity by adopting a dispersion normalization method to obtain a transformer online monitoring state quantity matrix X:
Figure BDA0001662412750000021
wherein, X1、X2And XrRespectively representing the online monitoring state quantity of the 1 st type transformer, the online monitoring state quantity of the 2 nd type transformer and the online monitoring state quantity of the r th type transformer, wherein subscripts 1 and 2 … … n represent time sequences;
(3) constructing a prediction model based on a grid long-short-term memory network and determining the structure of the prediction model, wherein the prediction model comprises a feature extractor and a feedforward neural network layer, and the feature extractor comprises N layers of grid long-short-term memory network layers; the structure of the prediction model comprises the number of grid long-time memory network layers, the number of neuron nodes of a feedforward neural network layer and each grid long-time memory network layer;
(4) taking a part of transformer online monitoring state quantities in the transformer online monitoring state quantity matrix X as input, and training a prediction model;
(5) inputting the other part of the transformer online monitoring state quantity in the transformer online monitoring state quantity matrix X as input data into a trained prediction model, and inputting the characteristic parameters between the input data and a prediction result into the feedforward neural network layer after the characteristic extractor extracts the characteristic parameters;
(6) and the feedforward neural network layer outputs a prediction result.
For the method for predicting the transformer state parameter trend, in the step (1), the acquired transformer on-line monitoring state quantity can acquire a plurality of groups of data according to factors influencing faults, such as: in some embodiments, the transformer on-line monitoring state quantity comprises transformer state data, power grid operation data related to the transformer and transformer substation environment meteorological data, so that the complex situation of mixing of multiple factors can be considered when the transformer state parameter trend prediction method is used for prediction, and compared with the prior art that only one or a few factors influencing faults can be considered, the prediction result obtained by the transformer state parameter trend prediction method is more accurate. And preferably the acquisition is time-sequentially acquired so that the data forms a time series matrix.
In the step (2), in order to improve the accuracy of the prediction model and to ensure that the accuracy of the prediction result finally obtained by the transformer state parameter trend prediction method is high, a dispersion standardization method is adopted to carry out normalization processing on the transformer online monitoring state quantity.
It should be noted that, in the technical solution of the present invention, the transformer online monitoring state quantity matrix X:
Figure BDA0001662412750000031
wherein, X1、X2And XrRespectively indicate the 1 st type transformer online monitoring state quantity, the 2 nd type transformer online monitoring state quantity and the r th type transformer online monitoring state quantity, and subscripts 1 and 2 … … n indicate time series.
For example: in some embodiments, the transformer on-line monitoring state quantity comprises transformer state data, grid operation data related to the transformer, and substation environment meteorological data, then X1、X2And XrRespectively representing transformer status data, grid operating data relating to the transformer and substation ambient meteorological data, while the transformer on-line monitoring state quantities include transformer status data over the past thirty days, grid operating data relating to the transformer and substation ambient meteorological data, the subscripts 1,2 … … n (n-30) representing a time series, and
Figure BDA0001662412750000032
and the transformer online monitoring state quantity represents the transformer state on the first day.
In the scheme, the transformer on-line monitoring state quantity is taken as an input characteristic quantity, the correlation of the self time dimension of each transformer on-line monitoring state quantity as input data is determined through characteristic extraction of Grid Long Short-term Memory (GLSTM) according to sequence data statistical relationship, abstract correlation characteristics need to be extracted from the transformer on-line monitoring state quantity as a high-dimensional complex parameter, strong correlation influence factors are activated to inhibit and weaken irrelevant and redundant information, meanwhile, the correlation relationship between the transformer on-line monitoring state quantities obtained by mining is taken as prior knowledge to carry out self-adaptive adjustment on the weight parameters of a prediction model (namely, the prediction model is trained), finally, the mined characteristic parameters are provided for a feedforward neural network layer to predict and obtain the change trend of the transformer state quantity, and finally, outputting a prediction result by a feedforward neural network layer.
The transformer state parameter trend prediction method can accurately predict the transformer state parameter trend, namely predict the transformer state parameter trend for a long time in the future, and therefore has practical guiding significance for the operation of a power grid.
Further, in the transformer state parameter trend prediction method of the present invention, the transformer on-line monitoring state quantity includes at least one of transformer state data, power grid operation data related to the transformer, and substation environment meteorological data.
Still further, in the method for predicting the trend of the state parameter of the transformer according to the present invention, the transformer state data includes at least one of a content of a gas dissolved in the transformer oil, a transformer oil temperature, and a transformer ground current; and/or the grid operating data relating to the transformer comprises at least one of operating current, active power and reactive power; and/or the substation ambient meteorological data comprises at least one of air temperature, ground temperature, relative humidity, and average wind speed.
Further, in the transformer state parameter trend prediction method, in the step (4), a prediction model is trained by adopting a time-based back propagation algorithm.
Further, in the transformer state parameter trend prediction method, in the step (3), the structure of the prediction model is determined by adopting an experimental method.
Accordingly, another objective of the present invention is to provide a transformer state parameter trend prediction system, which can predict a plurality of transformer state parameters of a transformer state, and the transformer state parameter trend prediction system can understand the influence of a plurality of transformer fault factors on the transformer state, and further understand the correlation between the factors, so that the transformer state parameter trend prediction system finally obtains a prediction result with high accuracy.
Based on the above purpose, the present invention provides a transformer state parameter trend prediction system, which includes:
the data acquisition module is used for acquiring the on-line monitoring state quantity of the transformer;
the data preprocessing module is used for carrying out normalization processing on the transformer online monitoring state quantity by adopting a dispersion standardization method to obtain a transformer online monitoring state quantity matrix X:
Figure BDA0001662412750000041
wherein, X1、X2And XrRespectively representing the online monitoring state quantity of the 1 st type transformer, the online monitoring state quantity of the 2 nd type transformer and the online monitoring state quantity of the r th type transformer, wherein subscripts 1 and 2 … … n represent time sequences;
the prediction model based on the grid long-short-term memory network comprises a feature extractor and a feedforward neural network layer, wherein the feature extractor comprises N layers of grid long-short-term memory network layers; and inputting data in the transformer online monitoring state quantity matrix X by the prediction model, and outputting a prediction result.
Further, in the transformer state parameter trend prediction system of the present invention, the transformer on-line monitoring state quantity includes at least one of transformer state data, grid operation data related to the transformer, and substation environment meteorological data.
Further, in the system for predicting the trend of the state parameter of the transformer, the state data of the transformer comprises at least one of the content of gas dissolved in transformer oil, the temperature of the transformer oil and the grounding current of the transformer; and/or the grid operating data relating to the transformer comprises at least one of operating current, active power and reactive power; and/or the substation ambient meteorological data comprises at least one of air temperature, ground temperature, relative humidity, and average wind speed.
Further, in the system for predicting the transformer state parameter trend, the feature extractor comprises 4 layers of grid long-time and short-time memory network layers.
According to the transformer state parameter trend prediction method, the incidence relation among various transformer state parameters (such as transformer state data, power grid operation data, transformer substation environment meteorological information and other transformer state parameters) is extracted by utilizing a grid long-term memory network, and the trend relation between the transformer state parameters and the future state quantity (namely the required prediction result) is excavated, so that the complex situation of various factors mixed can be considered when the transformer state parameter trend prediction method is used for prediction, and compared with the prior art that only a single or few factors influencing faults can be considered, the prediction result obtained by the transformer state parameter trend prediction method is more accurate.
In addition, the transformer state parameter trend prediction system also has the advantages.
Drawings
Fig. 1 is a schematic diagram of a transformer state parameter trend prediction system according to an embodiment of the present invention.
Fig. 2 is a schematic flow chart of a transformer state parameter trend prediction method according to an embodiment of the present invention.
Fig. 3 shows the recorded condition of the running current as one of the on-line monitoring state quantities of the transformer in a certain period of time.
FIG. 4 shows H2The concentration is used as one of the on-line monitoring state quantities of the transformer to record the condition in a certain time period.
Fig. 5 shows the CO concentration recorded as one of the on-line monitoring state quantities of the transformer over a certain period of time.
FIG. 6 shows CH4Concentration as transformer on-line monitoring state quantityOf the recording of one of the events over a period of time.
FIG. 7 shows C2H4The concentration is used as one of the recording conditions of the on-line monitoring state quantity of the transformer in a certain time period.
FIG. 8 shows C2H2The concentration is used as one of the recording conditions of the on-line monitoring state quantity of the transformer in a certain time period.
Fig. 9 shows the total hydrocarbon concentration as one of the on-line monitoring state quantities of the transformer recorded over a certain period of time.
Fig. 10 shows the recorded condition of the air temperature as one of the on-line monitoring state quantities of the transformer in a certain period of time.
Fig. 11 shows the top oil temperature as one of the online monitoring state quantities of the transformer, which is recorded in a certain period of time.
Fig. 12 shows a comparison between one of the online monitoring state quantities of the transformer using the top oil temperature as the top oil temperature in the transformer state parameter trend prediction method according to the present invention and the actually measured result.
Fig. 13 shows the error condition of fig. 12.
FIG. 14 shows a diagram H in the transformer state parameter trend prediction method according to the present invention2The concentration is used as an error condition after comparison of one of the online monitoring state quantities of the transformer and an actually measured result.
Fig. 15 shows the error after comparing one of the on-line monitoring state quantities of the transformer with the total hydrocarbon concentration according to the method for predicting the state parameter trend of the transformer in the invention with the actually measured result.
Detailed Description
The method and system for predicting the transformer state parameter trend according to the present invention will be further explained and illustrated with reference to the drawings and specific embodiments of the specification, however, the explanation and illustration should not be construed as an undue limitation on the technical solution of the present invention.
Fig. 1 is a schematic diagram of a transformer state parameter trend prediction system according to an embodiment of the present invention.
As shown in fig. 1, in the present technical solution, a transformer state parameter trend prediction system includes a data acquisition module, a data preprocessing module, and a prediction model based on a grid long-and-short term memory network, where the prediction model includes a feature extractor and a layer of feedforward neural network, and the feature extractor includes four layers of grid long-and-short term memory network layers.
It is needless to say that the number of the grid long-short term memory network layers may be set according to the specific situation of each embodiment, and in other embodiments, the number of the grid long-short term memory network layers may be five or other numbers.
Fig. 2 is a schematic flow chart of a transformer state parameter trend prediction method according to an embodiment of the present invention.
When the transformer state parameter trend prediction system is adopted to predict the transformer state parameter trend of the transformer state parameter, the prediction method comprises the following steps:
(1) acquiring transformer online monitoring state quantities through a data acquisition module, wherein the transformer online monitoring state quantities comprise transformer state data (such as the content of gas dissolved in transformer oil, the transformer oil temperature and the transformer grounding current), power grid operation data (such as operation current, active power and reactive power) related to a transformer and transformer substation environment meteorological data (such as air temperature, ground temperature, relative humidity and average wind speed);
(2) the online monitoring state quantity of the transformer is normalized by a data preprocessing module by adopting a dispersion standardization method to obtain an online monitoring state quantity matrix X of the transformer:
Figure BDA0001662412750000071
wherein, X1、X2And XrRespectively showing the on-line monitoring state quantity of the 1 st transformer, the on-line monitoring state quantity of the 2 nd transformer and the on-line monitoring state quantity of the r th transformer1,2 … … n represents a time series;
(3) constructing a prediction model based on a grid long-time memory network, and determining the structure prediction of the prediction model by an experimental method;
(4) taking a part of transformer online monitoring state quantities in a transformer online monitoring state quantity matrix X as input, and training a prediction model based on a time-based back propagation algorithm, wherein the structure of the prediction model comprises the number of grid long-and-short term memory network layers and the number of neuron nodes of a feedforward neural network layer and each grid long-and-short term memory network layer;
(5) inputting the other part of the transformer online monitoring state quantity in the transformer online monitoring state quantity matrix X as input data into a trained prediction model, and inputting the characteristic parameters between the input data and a prediction result into a feedforward neural network layer after extracting characteristic parameters by a characteristic extractor;
(6) and outputting the prediction result by the feedforward neural network layer.
The grid long-time memory network is provided with three gate control switches (an input gate, a forgetting gate and an output gate) to protect and control the state of the unit, and the input gate, the forgetting gate and the output gate respectively correspond to the injection, accumulation and output operations of the on-line monitoring state quantity of the transformer. The gate control switch realizes a time memory function, prevents gradient disappearance, and provides guarantee for deep mining of the association relation between the on-line monitoring state quantities of all transformers by the depth space information extraction function of the memory network layer when the grids are short.
In order to verify the accuracy of the prediction result of the transformer state parameter trend prediction method, the historical data of a certain transformer substation is used for prediction analysis.
Firstly, acquiring on-line monitoring state quantities of all transformers of the substation from 21/2010 to 28/2013 for 1200 days, and fig. 3 to 11 show recording conditions of different transformer on-line monitoring state quantities within 1200 days, for example: FIG. 3 shows a record of the operating current as one of the on-line monitoring state quantities of the transformer; FIG. 4 shows H2The concentration is used as the recording condition of one of the on-line monitoring state quantities of the transformer; FIG. 5 showsShowing the record condition of CO concentration as one of the on-line monitoring state quantities of the transformer; FIG. 6 shows CH4The concentration is used as the recording condition of one of the on-line monitoring state quantities of the transformer; FIG. 7 shows C2H4The concentration is used as the recording condition of one of the on-line monitoring state quantities of the transformer; FIG. 8 shows C2H2The concentration is used as the recording condition of one of the on-line monitoring state quantities of the transformer; FIG. 9 shows a record of total hydrocarbon concentration as one of the on-line monitoring state quantities of the transformer; fig. 10 shows a recording of air temperature as one of the on-line monitoring state quantities of the transformer; fig. 11 shows the top oil temperature as a record of one of the on-line monitored state quantities of the transformer.
Compared with the previous moments, the current state of the transformer state parameter has no great mutation, so that the long-term prediction has more practical guiding significance than the short-term prediction, and therefore, the prediction of the transformer state parameter is set according to the scheme, and the prediction is carried out by using state data of 60 days after 6 months and 28 days.
Taking the top oil temperature as the online monitoring state quantity of the transformer as an example, considering that all weights in the grid time interval memory network are randomly and initially distributed uniformly according to [ -0.05,0.05], all bias terms are initialized to 0, therefore, the momentum term in the initial training period (the first training period) is set to 0, and in the subsequent training period, the momentum term is set to 0.8, in addition, considering that in order to prevent the over-fitting situation, the signal loss rate is set to 0.2, the signal loss rate is carried out by adopting a back propagation algorithm along time, the model training degree is measured by adopting an average absolute error standard, finally, the network of the prediction model is determined to be 5 layers through an experimental method, and from each grid time interval memory network layer to the feedforward neural network layer, the number of neuron nodes in each layer is 297, 1200, 750, 200-60 in turn.
And the neural network structure of the single transformer state parameter obtained by long-term and short-term network memory without considering the correlation among the transformer state parameters, wherein the neuron nodes from the input layer to the output layer are 33-100-200-60.
Fig. 12 shows a comparison between one of the online monitoring state quantities of the transformer using the top oil temperature as the top oil temperature in the transformer state parameter trend prediction method according to the present invention and the actually measured result.
As shown in fig. 12, a curve I represents an actual top-layer oil temperature record obtained by actual measurement, a curve II represents a top-layer oil temperature prediction result obtained by using the transformer state parameter trend prediction method of the present application, and a curve III represents a top-layer oil temperature prediction result obtained by a neural network of a single transformer state parameter by long-and-short term network memory without considering the correlation among the transformer state parameters. As can be seen from fig. 12, the curve II fits more closely to the curve I than the curve III fits closely to the curve I.
Fig. 13 shows the error condition of fig. 12. As shown in fig. 13, a curve IV shows a prediction error of a top layer oil temperature prediction result obtained by the transformer state parameter trend prediction method of the present invention, and a curve V shows a prediction error of a top layer oil temperature prediction result obtained by the neural network of a single transformer state parameter through long and short term network memory without considering the correlation between the transformer state parameters, and as can be seen from fig. 13, an error of a top layer oil temperature prediction result obtained by the transformer state parameter trend prediction method of the present invention is smaller than an error of a top layer oil temperature prediction result obtained by the neural network of a single transformer state parameter through long and short term network memory without considering the correlation between the transformer state parameters, and the maximum prediction error is reduced from 20% to 10%.
As can be seen from fig. 12 and 13, by extracting the correlation between the state parameters of the transformers, the transformer state parameter trend prediction method of the present disclosure can better track the data change trend of the state parameters of the transformers, so that the finally obtained prediction effect has stronger robustness.
Similarly, the transformer state parameter trend prediction method and the long-term and short-term network memory are adopted to respectively store H by considering the incidence relation among the transformer state parameters2Concentrations and total hydrocarbon concentrations were predicted 60 days after 28 days 6 months.
FIG. 14 shows a diagram H in the transformer state parameter trend prediction method according to the present invention2The concentration is used as an error condition after comparison of one of the online monitoring state quantities of the transformer and an actually measured result.
As shown in FIG. 14, curve VI shows the curve obtained by the transformer state parameter trend prediction method of the present invention for H2The curve VII shows the neural network pair H of the single transformer state parameter obtained by long-term and short-term network memory without considering the correlation between the state parameters of the individual transformers2The prediction error of the concentration prediction result can be seen from fig. 14, and the prediction error of the concentration prediction result obtained by the transformer state parameter trend prediction method of the scheme is used for H2Comparing the error of the predicted result of the concentration with the neural network pair H of the single transformer state parameter obtained by long-short term network memory without considering the correlation between the state parameters of the individual transformers2The error of the predicted result of the concentration is smaller.
Fig. 15 shows the error after comparing one of the on-line monitoring state quantities of the transformer with the total hydrocarbon concentration according to the method for predicting the state parameter trend of the transformer in the invention with the actually measured result.
As shown in fig. 15, a curve VIII shows a prediction error of a prediction result of the total hydrocarbon concentration obtained by the transformer state parameter trend prediction method of the present invention, and a curve IX shows a prediction error of a prediction result of the total hydrocarbon concentration obtained by the neural network of the single transformer state parameter obtained by long and short term network memory without considering the correlation between the respective transformer state parameters, and it can be seen from fig. 15 that an error of a prediction result of the total hydrocarbon concentration obtained by the transformer state parameter trend prediction method of the present invention is smaller than an error of a prediction result of the total hydrocarbon concentration obtained by the neural network of the single transformer state parameter obtained by long and short term network memory without considering the correlation between the respective transformer state parameters.
It can be seen from fig. 14 and fig. 15 that the transformer state parameter trend prediction method of the present disclosure also has an effect of improving the accuracy of the prediction result of the content of the gas dissolved in the transformer oil.
In order to further verify the prediction effect of the prediction model constructed by the method compared with the conventional prediction models obtained by other conventional methods, the prediction is carried out on the oil temperature of the top layer 60 days after 6 months and 28 days, and error calculation is carried out on the result and the final actually measured result, and the comparison result is listed in table 1.
Table 1.
Model (model) GLSTM model AR model RBFNN model SVR model GM model
Mean error 9.87±2.21 >50 28.92±8.39 24.72±7.56 16.63±3.44
Note: in table 1, the GLSTM model refers to a prediction model obtained by the transformer state parameter trend prediction method in the present case, the AR model refers to a prediction model obtained by an autoregressive method, the RBFNN model refers to a prediction model obtained by a radial basis function neural network method, the SVR model refers to a prediction model obtained by a support vector machine regression method, and the GM model refers to a prediction model obtained by a multi-parameter gray method.
As can be seen from table 1, the correlation between the state parameters of the transformers is considered, so that the GLSTM model in the present application has a significantly reduced error and a smaller error fluctuation range compared to a prediction model without considering the correlation. In addition, compared with a GM model, the GLSTM model has comprehensive coverage information of the transformer state parameters, more complete extracted correlation characteristics, relatively small errors and no subjective judgment of a threshold value in the prediction process, namely the transformer state parameter trend prediction method takes the influence of various factors related to the transformer into consideration through the correlation among the transformer state parameters, thereby correcting the prediction model parameters and effectively reducing the transformer state parameter trend prediction errors.
It should be noted that the prior art in the protection scope of the present invention is not limited to the examples given in the present application, and all the prior art which is not inconsistent with the technical scheme of the present invention, including but not limited to the prior patent documents, the prior publications and the like, can be included in the protection scope of the present invention.
In addition, the combination of the features in the present application is not limited to the combination described in the claims of the present application or the combination described in the embodiments, and all the features described in the present application may be freely combined or combined in any manner unless contradictory to each other.
It should also be noted that the above-mentioned embodiments are only specific embodiments of the present invention. It is apparent that the present invention is not limited to the above embodiments and similar changes or modifications can be easily made by those skilled in the art from the disclosure of the present invention and shall fall within the scope of the present invention.

Claims (9)

1. A transformer state parameter trend prediction method is characterized by comprising the following steps:
(1) collecting the on-line monitoring state quantity of the transformer;
(2) carrying out normalization processing on the transformer online monitoring state quantity by adopting a dispersion normalization method to obtain a transformer online monitoring state quantity matrix X:
Figure FDA0002918826060000011
wherein, X1、X2And XrRespectively representing the online monitoring state quantity of the 1 st type transformer, the online monitoring state quantity of the 2 nd type transformer and the online monitoring state quantity of the r th type transformer, wherein subscripts 1 and 2 … … n represent time sequences;
(3) constructing a prediction model based on a grid long-short-term memory network and determining the structure of the prediction model, wherein the prediction model comprises a feature extractor and a feedforward neural network layer, and the feature extractor comprises N layers of grid long-short-term memory network layers; the structure of the prediction model comprises the number of grid long-time memory network layers, the number of neuron nodes of a feedforward neural network layer and each grid long-time memory network layer;
(4) taking a part of transformer online monitoring state quantities in the transformer online monitoring state quantity matrix X as input, and training a prediction model;
(5) inputting the other part of the transformer online monitoring state quantity in the transformer online monitoring state quantity matrix X as input data into a trained prediction model, and inputting the characteristic parameters between the input data and a prediction result into the feedforward neural network layer after the characteristic extractor extracts the characteristic parameters;
(6) and the feedforward neural network layer outputs a prediction result.
2. The transformer state parameter trend prediction method of claim 1, wherein the transformer on-line monitoring state quantity comprises at least one of transformer state data, grid operating data related to the transformer, and substation environmental meteorological data.
3. The transformer state parameter trend prediction method of claim 2, wherein the transformer state data comprises at least one of a content of dissolved gases in transformer oil, transformer oil temperature, transformer ground current; and/or the grid operating data relating to the transformer comprises at least one of operating current, active power and reactive power; and/or the substation ambient meteorological data comprises at least one of air temperature, ground temperature, relative humidity, and average wind speed.
4. The transformer state parameter trend prediction method of claim 1, wherein in step (4), the prediction model is trained using a time-based back propagation algorithm.
5. The transformer state parameter trend prediction method of claim 1, wherein in step (3), the structure of the prediction model is determined experimentally.
6. A transformer state parameter trend prediction system, comprising:
the data acquisition module is used for acquiring the on-line monitoring state quantity of the transformer;
the data preprocessing module is used for carrying out normalization processing on the transformer online monitoring state quantity by adopting a dispersion standardization method to obtain a transformer online monitoring state quantity matrix X:
Figure FDA0002918826060000021
wherein, X1、X2And XrRespectively representing the online monitoring state quantity of the 1 st type transformer, the online monitoring state quantity of the 2 nd type transformer and the online monitoring state quantity of the r th type transformer, wherein subscripts 1 and 2 … … n represent time sequences;
the prediction model based on the grid long-short-term memory network comprises a feature extractor and a feedforward neural network layer, wherein the feature extractor comprises N layers of grid long-short-term memory network layers; and inputting data in the transformer online monitoring state quantity matrix X by the prediction model, and outputting a prediction result.
7. The transformer state parameter trend prediction system of claim 6, wherein the transformer on-line monitoring state quantity comprises at least one of transformer state data, grid operating data related to the transformer, and substation environmental meteorological data.
8. The transformer state parameter trend prediction system of claim 7, wherein the transformer state data includes at least one of a content of dissolved gases in transformer oil, transformer oil temperature, transformer ground current; and/or the grid operating data relating to the transformer comprises at least one of operating current, active power and reactive power; and/or the substation ambient meteorological data comprises at least one of air temperature, ground temperature, relative humidity, and average wind speed.
9. The transformer state parameter trend prediction system of claim 6, wherein the feature extractor comprises 4 layers of grid long-time memory network layers.
CN201810468349.6A 2018-05-16 2018-05-16 Method and system for predicting state parameter trend of transformer Active CN108390380B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810468349.6A CN108390380B (en) 2018-05-16 2018-05-16 Method and system for predicting state parameter trend of transformer

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810468349.6A CN108390380B (en) 2018-05-16 2018-05-16 Method and system for predicting state parameter trend of transformer

Publications (2)

Publication Number Publication Date
CN108390380A CN108390380A (en) 2018-08-10
CN108390380B true CN108390380B (en) 2021-04-06

Family

ID=63071092

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810468349.6A Active CN108390380B (en) 2018-05-16 2018-05-16 Method and system for predicting state parameter trend of transformer

Country Status (1)

Country Link
CN (1) CN108390380B (en)

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110045237A (en) * 2019-04-08 2019-07-23 国网上海市电力公司 Transformer state parametric data prediction technique and system based on drosophila algorithm optimization
CN110045236A (en) * 2019-04-08 2019-07-23 国网上海市电力公司 Transformer state parametric data prediction technique and system based on core pivot element analysis optimization
CN111611422B (en) * 2020-05-21 2020-12-01 广东省地震局 SVC-based method and system for automatically generating qualitative graph in earthquake disaster risk assessment
CN112036601B (en) * 2020-07-23 2023-10-27 国网江苏省电力有限公司检修分公司 Oil temperature prediction method and system for synchronous phase modulation engine oil system
DE102022105138A1 (en) * 2022-03-04 2023-09-07 Dehn Se Method and system for monitoring a power grid, method for training artificial intelligence for forecasting a future state of a power grid, computer program and computer-readable data carrier

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103678765A (en) * 2013-10-31 2014-03-26 上海交通大学 Transformer operating state comprehensive evaluation method based on on-line monitoring
CN108037378A (en) * 2017-10-26 2018-05-15 上海交通大学 Running state of transformer Forecasting Methodology and system based on long memory network in short-term

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR101980603B1 (en) * 2016-05-20 2019-05-22 구글 엘엘씨 Relating to predicting the motion (s) of the object (s) in the robotic environment based on the image (s) capturing the object (s) and parameter (s) for future robot motion in the environment Methods and apparatus

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103678765A (en) * 2013-10-31 2014-03-26 上海交通大学 Transformer operating state comprehensive evaluation method based on on-line monitoring
CN108037378A (en) * 2017-10-26 2018-05-15 上海交通大学 Running state of transformer Forecasting Methodology and system based on long memory network in short-term

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
基于深度信念网络的变压器油中溶解气体浓度预测方法;代杰杰 等;《电网技术》;20170831;第41卷(第8期);第2737-2742页 *
采用LSTM 网络的电力变压器运行状态预测方法研究;代杰杰 等;《高电压技术》;20180430;第44卷(第4期);第1099-1106页 *

Also Published As

Publication number Publication date
CN108390380A (en) 2018-08-10

Similar Documents

Publication Publication Date Title
CN108390380B (en) Method and system for predicting state parameter trend of transformer
CN108037378B (en) Transformer operation state prediction method and system based on long-time and short-time memory network
CN111237134B (en) Offshore double-fed wind driven generator fault diagnosis method based on GRA-LSTM-stacking model
CN116757534A (en) Intelligent refrigerator reliability analysis method based on neural training network
Fu et al. A re-optimized deep auto-encoder for gas turbine unsupervised anomaly detection
Pandit et al. Data‐driven weather forecasting models performance comparison for improving offshore wind turbine availability and maintenance
CN112818604A (en) Wind turbine generator risk degree assessment method based on wind power prediction
Xiao et al. Long-term predictive opportunistic replacement optimisation for a small multi-component system using partial condition monitoring data to date
CN114397526A (en) Power transformer fault prediction method and system driven by state holographic sensing data
CN110737975B (en) Wind power plant wind speed and power prediction and abnormality correction method based on empirical mode decomposition and autoregressive model
CN110110784B (en) Transformer fault identification method based on transformer related operation data
CN116793666A (en) Wind turbine generator system gearbox fault diagnosis method based on LSTM-MLP-LSGAN model
CN108282360B (en) Fault detection method for long-term and short-term prediction fusion
CN116521406A (en) Method for detecting anomaly of non-overrun flight parameter data of aero-engine based on residual gate GRU-VAE model
CN114049014A (en) Method, device and system for evaluating operation state of offshore wind turbine generator
Mesa-Jiménez et al. Early warning signals of failures in building management systems
CN115578084A (en) Wind turbine generator set frequency converter fault early warning method based on deep convolution self-encoder
Babaali et al. Weighted ensemble learning for real‐time short‐term voltage stability assessment with phasor measurements data
Shi et al. Phase I analysis of hidden operating status for wind turbine
CN117113886B (en) Pressure prediction method and device
Shen Classification of Wind Turbine Blade Performance State through Statistical Methods
Liu et al. Ultra-short-term wind power forecasting based on stacking model
Seshan et al. LSTM-based autoencoder models for real-time quality control of wastewater treatment sensor data
Zhao et al. An interpretable ultra-short-term wind power prediction model based on the feature matrix reconstruction through regression trees
Xiao et al. Multi-Sensor Infusion and Data-Physics Model Based Remaining Life Prediction

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
GR01 Patent grant
GR01 Patent grant