CN117672403A - Method for predicting concentration trend of dissolved gas in power transformer oil - Google Patents
Method for predicting concentration trend of dissolved gas in power transformer oil Download PDFInfo
- Publication number
- CN117672403A CN117672403A CN202311582300.0A CN202311582300A CN117672403A CN 117672403 A CN117672403 A CN 117672403A CN 202311582300 A CN202311582300 A CN 202311582300A CN 117672403 A CN117672403 A CN 117672403A
- Authority
- CN
- China
- Prior art keywords
- dissolved gas
- concentration
- data
- missing
- power transformer
- 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 28
- 238000012549 training Methods 0.000 claims abstract description 22
- 238000004140 cleaning Methods 0.000 claims abstract description 8
- 238000004587 chromatography analysis Methods 0.000 claims description 10
- JDDHUROHDHPVIO-UHFFFAOYSA-N Piperazine citrate Chemical compound C1CNCCN1.C1CNCCN1.C1CNCCN1.OC(=O)CC(O)(C(O)=O)CC(O)=O.OC(=O)CC(O)(C(O)=O)CC(O)=O JDDHUROHDHPVIO-UHFFFAOYSA-N 0.000 claims description 8
- 238000010586 diagram Methods 0.000 claims description 8
- 229940103147 propet Drugs 0.000 claims description 8
- 238000011208 chromatographic data Methods 0.000 claims description 7
- 238000012795 verification Methods 0.000 claims description 7
- 238000004458 analytical method Methods 0.000 claims description 6
- 238000005457 optimization Methods 0.000 claims description 2
- 238000007781 pre-processing Methods 0.000 claims description 2
- 238000001514 detection method Methods 0.000 abstract description 4
- 230000007774 longterm Effects 0.000 abstract description 3
- 239000007789 gas Substances 0.000 description 72
- 229910052739 hydrogen Inorganic materials 0.000 description 12
- 239000001257 hydrogen Substances 0.000 description 12
- 230000000737 periodic effect Effects 0.000 description 12
- CURLTUGMZLYLDI-UHFFFAOYSA-N Carbon dioxide Chemical compound O=C=O CURLTUGMZLYLDI-UHFFFAOYSA-N 0.000 description 10
- 230000002159 abnormal effect Effects 0.000 description 10
- VNWKTOKETHGBQD-UHFFFAOYSA-N methane Chemical compound C VNWKTOKETHGBQD-UHFFFAOYSA-N 0.000 description 10
- 101001095088 Homo sapiens Melanoma antigen preferentially expressed in tumors Proteins 0.000 description 8
- 102100037020 Melanoma antigen preferentially expressed in tumors Human genes 0.000 description 8
- 230000008859 change Effects 0.000 description 8
- 238000012217 deletion Methods 0.000 description 8
- 230000037430 deletion Effects 0.000 description 8
- 150000002431 hydrogen Chemical class 0.000 description 8
- 238000012360 testing method Methods 0.000 description 8
- UFHFLCQGNIYNRP-UHFFFAOYSA-N Hydrogen Chemical compound [H][H] UFHFLCQGNIYNRP-UHFFFAOYSA-N 0.000 description 7
- 230000000694 effects Effects 0.000 description 6
- 238000012544 monitoring process Methods 0.000 description 6
- 239000004215 Carbon black (E152) Substances 0.000 description 5
- UGFAIRIUMAVXCW-UHFFFAOYSA-N Carbon monoxide Chemical compound [O+]#[C-] UGFAIRIUMAVXCW-UHFFFAOYSA-N 0.000 description 5
- OTMSDBZUPAUEDD-UHFFFAOYSA-N Ethane Chemical compound CC OTMSDBZUPAUEDD-UHFFFAOYSA-N 0.000 description 5
- VGGSQFUCUMXWEO-UHFFFAOYSA-N Ethene Chemical compound C=C VGGSQFUCUMXWEO-UHFFFAOYSA-N 0.000 description 5
- 239000005977 Ethylene Substances 0.000 description 5
- HSFWRNGVRCDJHI-UHFFFAOYSA-N alpha-acetylene Natural products C#C HSFWRNGVRCDJHI-UHFFFAOYSA-N 0.000 description 5
- 229910002092 carbon dioxide Inorganic materials 0.000 description 5
- 239000001569 carbon dioxide Substances 0.000 description 5
- 229910002091 carbon monoxide Inorganic materials 0.000 description 5
- 125000002534 ethynyl group Chemical group [H]C#C* 0.000 description 5
- 238000011156 evaluation Methods 0.000 description 5
- 229930195733 hydrocarbon Natural products 0.000 description 5
- 150000002430 hydrocarbons Chemical class 0.000 description 5
- 238000012952 Resampling Methods 0.000 description 3
- 239000011159 matrix material Substances 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 230000001932 seasonal effect Effects 0.000 description 2
- 230000011218 segmentation Effects 0.000 description 2
- 238000007476 Maximum Likelihood Methods 0.000 description 1
- 230000005856 abnormality Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004040 coloring Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 238000004870 electrical engineering Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000013213 extrapolation Methods 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000009421 internal insulation Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003860 storage Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
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
- Housings And Mounting Of Transformers (AREA)
Abstract
The invention discloses a method for predicting concentration trend of dissolved gas in power transformer oil, which comprises the steps of carrying out original sequence pretreatment and outlier cleaning on the concentration of the dissolved gas in the power transformer oil to obtain model data of the concentration of the dissolved gas in the power transformer oil, and further constructing training set data; and establishing a Prophet prediction model, optimizing according to training set data to obtain an optimized Prophet prediction model, and predicting the concentration trend of the dissolved gas according to the optimized Prophet prediction model. The method can accurately predict the concentration of the dissolved gas in the power transformer oil with the missing value on line for a short term and/or a long term, and improves the operation and detection efficiency of the power transformer substation.
Description
Technical Field
The invention belongs to the technical field of electrical engineering, and relates to a method for predicting concentration trend of dissolved gas in power transformer oil.
Background
Because the state quantities of the power transformer such as the load, the oil temperature and the internal insulation development state in operation change constantly, the oil chromatographic data also fluctuates constantly, and the fluctuation is not linear and smooth, but has a certain trend and periodicity for a long time.
At present, the oil chromatography technology is a powerful means for judging the state of a power transformer, and trend prediction of oil chromatography data plays a very important role in early warning work of the state of the transformer. The existing oil chromatography online monitoring system is operated remotely on the equipment site for a long time, has no system function self-checking function and no early warning capability, so that online accurate prediction of the concentration trend of dissolved gas in the power transformer oil cannot be performed, and the improvement of the operation and detection efficiency of a transformer substation is limited; in addition, the existing data for predicting the concentration trend of the dissolved gas in the power transformer oil has high requirements, and if the sampled data has missing values or abnormal values, the prediction cannot be performed, so that the application of the method is further limited.
In view of the above, the present invention is needed to provide a method for accurately predicting the concentration trend of the dissolved gas in the power transformer oil, so as to effectively improve the operation and inspection efficiency of the transformer substation.
Disclosure of Invention
In order to overcome the problems, the inventor researches a method for predicting the concentration of the dissolved gas in the power transformer oil, researches a method for predicting the concentration trend of the dissolved gas in the power transformer oil, can accurately predict the concentration trend of the dissolved gas in the power transformer oil on line, and simultaneously effectively improves the operation and detection efficiency of a transformer substation, thereby completing the invention.
In particular, the invention aims to provide a method for predicting concentration trend of dissolved gas in power transformer oil, which comprises the following steps:
step 1, obtaining dissolved gas concentration model data in power transformer oil;
step 2, constructing training set data based on the dissolved gas concentration model data;
step 3, a propset prediction model is established;
step 4, fitting the Prophet prediction model according to the training set data to obtain an optimized Prophet prediction model;
and 5, predicting the concentration trend of the dissolved gas according to the optimized Prophet prediction model.
In the step 1, the concentration of the dissolved gas in the power transformer oil is subjected to original sequence pretreatment and abnormal value cleaning, and the model data of the concentration of the dissolved gas in the power transformer oil is obtained.
Wherein the preprocessing comprises: acquiring the original concentration of dissolved gas in the power transformer oil, performing missing verification, and marking missing unknown data, non-conforming data format and non-conforming historical data as missing values; then, timing verification is performed.
Wherein the outlier cleaning includes: the outlier is converted into a missing value.
In step 2, based on the dissolved gas concentration model data, performing a missing state analysis; and judging the missing characteristics of the oil chromatographic data according to the missing state, and further constructing training set data.
Wherein the deletion state analysis includes: the sequence occurrence missing value of the dissolved gas concentration model data is marked as 0, and the correlation of the time sequence of occurrence 0 is used as a missing thermodynamic diagram.
Wherein the missing features of the oil chromatography data include a completely random loss and a non-completely random loss.
In step 3, the Prophet predictive model is represented by trend fit, periodic fit, and uncertainty fit of dissolved gas concentrations.
In step 4, the Prophet prediction model is optimized by adjusting the trend parameter and the periodicity parameter.
In step 5, the optimized Prophet prediction model is used for carrying out time sequence extension prediction to obtain the gas concentration to be predicted, so as to realize the prediction of the concentration trend of the dissolved gas in the power transformer oil.
The invention has the beneficial effects that:
(1) The method for predicting the concentration trend of the dissolved gas in the power transformer oil can accurately predict the concentration of the dissolved gas in the power transformer oil with the missing value on line for a short term and/or a long term, and has good expansibility, interpretability and flexibility.
(2) The method for predicting the concentration trend of the dissolved gas in the power transformer oil provides a basis for early warning and diagnosis of the state of the power transformer, can realize tracking and predicting of the fault risk of the power transformer, and effectively improves the operation and detection efficiency of the power transformer substation.
Drawings
Various other advantages and benefits of the present invention will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The drawings are only for purposes of illustrating the preferred embodiments and are not to be construed as limiting the invention. It is evident that the figures described below are only some embodiments of the invention, from which other figures can be obtained without inventive effort for a person skilled in the art.
In the drawings:
FIG. 1 shows a state diagram of the absence of oil chromatography data in example 1;
FIG. 2 shows a missing thermodynamic diagram in example 1;
fig. 3 (a) to 3 (d) show index evaluation graphs of the observation area in example 1, in which fig. 3 (a) shows a mean absolute percentage error graph of the observation area, fig. 3 (b) shows a median error absolute value graph of the observation area, fig. 3 (c) shows a root mean square error graph of the observation area, and fig. 3 (d) shows an estimated coverage graph of the observation area;
fig. 4 (a) to 4 (b) show predicted result graphs with a loss rate of 20%, wherein fig. 4 (a) shows a fitted graph with a loss rate of 20%, and fig. 4 (b) shows trend and periodicity characteristics;
fig. 5 shows a comparison of the actual data and the predicted data of the concentration of dissolved hydrogen in example 2.
Detailed Description
Specific embodiments of the present invention will be described in more detail below with reference to fig. 1 to 5. While specific embodiments of the invention are shown in the drawings, it should be understood that the invention may be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.
It should be noted that certain terms are used throughout the description and claims to refer to particular components. Those of skill in the art will understand that a person may refer to the same component by different names. The description and claims do not identify differences in terms of components, but rather differences in terms of the functionality of the components. As used throughout the specification and claims, the terms "include" and "comprise" are used in an open-ended fashion, and thus should be interpreted to mean "include, but not limited to. The description hereinafter sets forth a preferred embodiment for practicing the invention, but is not intended to limit the scope of the invention, as the description proceeds with reference to the general principles of the description. The scope of the invention is defined by the appended claims.
For the purpose of facilitating an understanding of the embodiments of the present invention, reference will now be made to the drawings, by way of example, and specific examples of which are illustrated in the accompanying drawings.
The invention provides a method for predicting concentration trend of dissolved gas in power transformer oil, which comprises the following steps:
step 1, obtaining dissolved gas concentration model data in power transformer oil;
step 2, constructing training set data based on the dissolved gas concentration model data;
step 3, a propset prediction model is established;
step 4, optimizing the propet prediction model according to the training set data to obtain an optimized propet prediction model;
and 5, predicting the concentration trend of the dissolved gas according to the optimized Prophet prediction model.
The following describes a method for predicting the concentration trend of dissolved gas in power transformer oil in detail.
And step 1, obtaining the concentration model data of the dissolved gas in the power transformer oil.
According to a preferred embodiment, the concentration of the dissolved gas in the power transformer oil is subjected to original sequence pretreatment and abnormal value cleaning, and the model data of the concentration of the dissolved gas in the power transformer oil are obtained.
The dissolved gas in the power transformer oil comprises hydrogen, methane, ethylene, acetylene, ethane, carbon monoxide, carbon dioxide and total hydrocarbon.
According to the invention, the pretreatment comprises: resampling an original sequence of the concentration of the dissolved gas in the power transformer oil by taking 1-4 hours as a sampling period to obtain the original concentration of the dissolved gas in the power transformer oil; then, carrying out missing verification on the original concentration of the dissolved gas in the power transformer oil, and marking missing unknown data (null value), non-conforming data formats such as negative numbers and non-conforming historical data (namely data with errors in time) as missing values; then, a timing check is performed, i.e. the timeliness problem of the timing data is solved, and when the time index contains repeated time stamps, the first value is reserved.
According to the present invention, the outlier cleaning includes: the outlier is converted into a missing value.
Further, the abnormal value refers to an abnormal value of uploaded data or a false value generated in system storage due to sensor failure, and is distinguished from an abnormal occurrence of a real state quantity measured by a sensor, and the abnormal value comprises an outlier, a horizontal shift, a periodic abnormality and the like.
According to the invention, after pretreatment and abnormal value cleaning are carried out on an original sequence of concentration of dissolved gas in the power transformer oil, a sequence of only missing values and abnormal value-free values of 8 characteristic gases including hydrogen concentration, methane concentration, ethylene concentration, acetylene concentration, ethane concentration, carbon monoxide concentration, carbon dioxide concentration and total hydrocarbon concentration is obtained, and then the dissolved gas concentration model data in the power transformer oil is obtained.
And 2, constructing training set data based on the dissolved gas concentration model data.
In step 2, based on the dissolved gas concentration model data, performing a missing state analysis; and judging the missing characteristics of the oil chromatographic data according to the missing state, and further constructing training set data.
Wherein the deletion state analysis means: the sequence occurrence missing value of the dissolved gas concentration model data is marked as 0, and the correlation of the time sequence of occurrence 0 is used as a missing thermodynamic diagram, namely, a statistical chart of the data is displayed by coloring color blocks, wherein a blank part is the missing part of the data.
In the prior art, a great part of reasons for low online prediction quality of oil chromatography are high in the deletion rate from the deletion value. In order to study the influence of different loss rates on the prediction accuracy of the time series model, the present inventors first conducted a loss state observation on the dissolved gas concentration model data after pretreatment. In one embodiment, the missing state of the dissolved gas concentration model data is shown in fig. 1, the white part in the figure is the missing value, the missing value is marked as 0, and the correlation of the time sequence of 0 is used as the missing thermodynamic diagram, as shown in fig. 2. The missing characteristics of the oil chromatographic data are judged more reliably and accurately according to the missing state and the missing thermodynamic diagram.
In the present invention, the missing features of the oil chromatography data include a completely random loss and a non-completely random loss.
The complete random loss is understood to be other random loss besides the non-complete random loss, specifically, 8 sensors for monitoring the concentration of 8 gases dissolved in the power transformer oil (8 sensors are respectively used for monitoring 8 different concentrations of gases dissolved in the power transformer oil, namely, 1 sensor is used for detecting only one concentration of gases, or stated differently, 1 sensor is used for monitoring 1 concentration of gases dissolved in the power transformer oil), the observed values of all the observed values are completely random lost, all the row of missing values appear completely random on a matrix of a multi-element sequence, at this time, the missing values of the concentration of 8 gases dissolved in the power transformer oil are completely uncorrelated, and when the complete random loss occurs, trend prediction needs to be carried out on the 8 gas sequences dissolved in the power transformer oil respectively.
Further, the non-complete random loss refers to that each sensor loses an observed value within a plurality of consecutive days or hours, or that each sensor loses a value within a plurality of consecutive days or hours, at this time, the missing values of the dissolved 8 gas concentrations in the power transformer oil occur at the same time, that is, the dissolved 8 gas concentration data in the power transformer oil are simultaneously lost at the same time, and the frequencies of the missing values of each column of data appear as being related on a matrix of a plurality of sequences, so when the non-random loss occurs, 1 gas sequence can be randomly selected for trend prediction, for example, a hydrogen sequence is selected for trend prediction.
According to a preferred embodiment, 80% -90% of the time sequence of the dissolved gas concentration model data is constructed as training set data, and 10% -20% of the time sequence of the dissolved gas concentration model data is constructed as test set data; for example, the first 90% of the time series of dissolved gas concentration model data is constructed as training set data, and the last 10% of the time series of dissolved gas concentration model data is constructed as test set data.
In the invention, the test set data can be selected to verify the validity of the dissolved gas concentration trend prediction.
And 3, establishing a Prophet prediction model.
In step 3, the Prophet prediction model is represented by a trend term, a period term and an uncertainty term of the concentration of the dissolved gas, and the relationship is represented by the formula (1), and the relationship is used as the established Prophet prediction model, wherein the formula (1) is a function of an independent variable t, and is specifically represented as:
y (t) =g (t) +s (t) +h (t) formula (1);
wherein: y (t) is a concentration sequence of dissolved gas in the power transformer oil, g (t) is a trend term which simulates the aperiodic variation of time series values; h (t) is an uncertainty term that is extended forward by the generated uncertainty model to estimate uncertainty in the predicted trend; s (t) is a period term.
The trend item fitting Prophet prediction model by constructing a segmented logistic growth model is specifically as follows:
in the formula (2):
t is time;
c (t) is the bearing capacity, namely the upper limit that the trend term can reach;
k (t) is the growth rate, which is a piecewise function that varies with time;
m (t) is an offset parameter, and the trend function g (t) can be kept continuous by adjusting m for different segmentation points.
Since k (t) is time-varying, and this variation is not continuous. Thus, the concept of segmentation points is introduced in the model, one term k (t) being instead:
in the formula (3):
k represents the amount of change of the basis;
i represents a change point number;
S i representing the time corresponding to the change point sequence number;
d i representing the variation value of each time point k (t).
For d i Is first equi-frequency to produce a large number of candidate points, each sample of which is derived from a Laplacian distribution, i.e., d i Laplace (0, t), where the estimate of the parameter τ is denoted λ, denoted as the maximum likelihood estimate of the rate scale parameter (which is denoted as the trending parameter by the present invention).
Wherein the uncertainty term of the Prophet predictive model is calculated by constructing a trend generating model, i.e. the uncertainty of the predicted trend is measured by assuming that the average frequency and amplitude of the future rate change is the same as the history. In the prediction phase, when the model is extrapolated past history to make predictions, it is assumed that the future and history are uniformly distributed, so once λ is determined, this generated model is used to simulate possible future trends and consider the upper and lower limit ranges of all possible trends simulated as uncertainty intervals, specifically:
in the formula (4):
t is the number of predicted points;
j is the serial number of the predicted point;
s is the number of rate change points (segment points) in the predicted point.
δ j The rate for each change point is changed.
The method comprises the steps of fitting a periodic term of a Prophet prediction model by constructing a periodic effect model, wherein the periodic term is specifically as follows:
in formula (5):
n is fourier series, n=1, 2, … …, N;
a n ，b n fourier coefficients corresponding to different fourier series;
t is time;
p is a regular period of the desired time sequence;
β＝[a 1 ，b 1 ，...，a N ，b N ] T to fit the 2N parameters that periodically need to be estimated, β -Normal (0, σ) is used 2 ) The periodicity of the smoothing priors is set, where σ is the periodicity parameter.
And step 4, optimizing the propet prediction model according to the training set data to obtain an optimized propet prediction model.
In step 4, the data output by the propet prediction model (i.e. under the condition that all parameters from the formula (2) to the formula (5) in the step 3 are determined, the right side item of the formula (1) is summed, a training set data difference relation corresponding to a group of y (t) and t moment is obtained, a trend parameter lambda and a periodic parameter sigma are adjusted, and optimization of the propet prediction model is achieved.
Further, by adjusting the trend parameter lambda, the prediction trend is more flexible, and the strength adjustment of a sparse prior (the lambda determining process is the sparse prior, which means that an extrapolation change point in the prediction process is sparse in a prediction time period) is realized, so that the problem of over-fitting or under-fitting of the Prophet prediction model in the step 3 is solved; the degree to which the data is subject to seasonal fluctuations is characterized by adjusting the periodicity parameter sigma to control the flexibility of the seasonal features of the oil chromatography data.
According to a preferred embodiment, the trend parameter λ is set to 0.05; the periodicity parameter sigma is set to 10. Through super-parameter tuning of the propset prediction model, the periodic parameter sigma is traversed in the set {0.01,0.1,1.0,10.0}, and meanwhile the trend parameter lambda is traversed in the set {0.001,0.01,0.1}, wherein the root mean square error (root mean squared error, RMSE) is selected for the evaluation index of the model and used for evaluating the difference between the data (i.e. y (t)) output by the propset prediction model and the training set data, so as to obtain the optimal prediction model parameter, and the RMSE is expressed as:
wherein:
m is the total number of training sets;
y i is a true value;
the fitting value obtained according to the optimized prophet prediction model is the y value on the left side of the formula (1) which is adjusted once each time the trend parameter lambda and the periodicity parameter sigma are adjusted once corresponding to the t moment;
and taking the trend parameter lambda and the periodic parameter sigma corresponding to the minimum value of the RMSE as the trend parameter lambda and the periodic parameter sigma in the optimal prediction model.
And 5, predicting the concentration trend of the dissolved gas according to the optimized propset prediction model.
In step 5, the Prophet prediction model optimized in step 4 is used for carrying out time sequence extension prediction to obtain the gas concentration to be predicted, so as to realize the prediction of the concentration trend of the dissolved gas in the power transformer oil. Optionally, the test set data is used for verifying the validity of the trend of the concentration of the dissolved gas so as to realize the validity and the correctness of the trend prediction of the concentration data of other dissolved gases.
Further, the prediction results are compared with the test set data, and the accuracy and the applicability of the prediction results of the dissolved gas concentration propset prediction model in the optimized voltage transformer oil are evaluated through root mean square error (root mean squared error, RMSE), average absolute percentage error (mean absolute percent error, MAPE), median absolute value of error (median absolute percent error, MDAPE) and/or estimated coverage (coverage).
The root mean square error (root mean squared error, RMSE) is expressed as follows:
the mean absolute percentage error (mean absolute percent error, MAPE) is expressed as follows:
the median absolute error value (median absolute percent error, MDAPE) is expressed as follows:
in the formulas (7) to (9):
n is the total number of test sets;
y i is a true value;
is a predicted value;
median is the median.
Wherein, the smaller the values of RMSE and MAPE, the better the fitting effect of the training set is indicated.
Examples
The invention is further described below by means of specific examples, which are however only exemplary and do not constitute any limitation on the scope of protection of the invention.
Example 1
(1) And (3) resampling the original concentration of dissolved hydrogen, methane, ethylene, acetylene, ethane, carbon monoxide, carbon dioxide and total hydrocarbon (expressed by total in the attached drawing of the specification) gas in the running alternating current 220kV transformer oil of a 220kV transformer substation in a period of one hour in a certain month to obtain 871 groups of oil chromatographic sample data sets, respectively carrying out deletion verification and time sequence verification on the original concentrations of the hydrogen, methane, ethylene, acetylene, ethane, carbon monoxide, carbon dioxide and total hydrocarbon gas, and simultaneously converting the abnormal value into the missing value to obtain the 8 gas concentration model data dissolved in the power transformer oil.
(2) According to the 8 gas concentration model data, 8 gas concentration missing states are obtained, as shown in fig. 1, and the white part in fig. 1 is the missing value. The missing values appearing in the above 8 kinds of gas concentration model data are marked as 0, and the correlation of the time series of the appearance of 0 is used as a missing thermodynamic diagram, and the results are shown in fig. 2.
As can be seen from fig. 1 and fig. 2, the foregoing 871 set of oil chromatographic sample data sets is characterized by a complete correlation between the frequencies of the missing values of the multi-element sequence matrix, and belongs to a non-complete random loss, so that the trend prediction of the concentration of dissolved gas in the power transformer oil can be completely simulated by constructing a single sequence of missing.
The trend prediction is performed by a hydrogen sequence, the time span is from 0:00:00 of 10 months of 2018 to 24:00:00 of 30 months of 2018, and the sampling is performed again by taking 1 hour as a period, so that 720 real observation points are obtained. The first 90% (first 27 days, total 648) of the hydrogen resampled dataset were used as training set data, and the last 10% (last 3 days, total 72) of the hydrogen resampled dataset were used as test set data.
(3) And (3) establishing a Prophet prediction model according to the formulas (1) to (5).
(4) Setting the trend parameter lambda as 0.05, setting the periodicity parameter sigma as 10, and optimizing the Prophet prediction model through training set data.
The periodic parameter sigma is traversed in a set {0.01,0.1,1.0,10.0}, meanwhile, the trend parameter lambda is traversed in a set {0.001,0.01 and 0.1}, and the accuracy and the applicability of the prediction result of the optimized propset prediction model are evaluated by taking the RMSE as an evaluation index, wherein the result is shown in a table 1.
Table 1:
as can be seen from table 1, the optimized propset prediction model shows a prominent set of parameters of λ=0.01 and σ=1.0, and thus this set of parameters is selected as the optimal prediction model parameters.
(5) And (3) taking the time sequence extension of the Prophet prediction model optimized in the step (4) as an output quantity to obtain a hydrogen prediction result.
Hydrogen, 10 month 18 day 0:00:00, 10 month 19 day 0:00, 2018, 10 month 20 day 0:00:00 were selected as cut-off points, and for each cut-off point, only data prior to the cut-off point was used to fit the model. The predicted value is then compared with the actual value. MAPE, MDAPE, RMSE and estimated coverage (coverage) are used as evaluation indicators, compared to the field of view of the observation area, which spans 20 days. Each evaluation index is shown in fig. 3 (a) to 3 (d), where fig. 3 (a) shows MAPE, fig. 3 (b) shows MDAPE, fig. 3 (c) shows RMSE, and fig. 3 (d) shows estimated coverage (coverage).
As can be seen from fig. 3 (a) to fig. 3 (d), the optimized propset prediction model predicts less than 1% MAPE for the data set for the next 1 day, and increases to more than 4% for the next ten days; the estimated coverage increases with increasing observation area, the longer the prediction time, the greater the estimated coverage, which is very helpful for long-term efficient prediction, providing an uncertainty range for future trends.
(6) And constructing fitting effect experiments of the missing values under the non-random missing states with different missing rates by taking the hydrogen resampling data as standard time sequence data.
Based on the sequence with 100% of integrity and 720 observation points, the non-complete random deletion sequences with 10%, 20%, 30%, 40% and 50% deletion rate are respectively constructed according to the deletion characteristics of the oil chromatography on-line monitoring data extracted in the step (2). Fitting under a Prophet prediction model is carried out on the model, and the prediction time length is set to be 24 hours. The fitting effect of the missing parts of the optimized propset prediction model was evaluated with RMSE and MAPE indices and the results are shown in table 2.
Table 2:
| loss rate | RMSE | MAPE |
| 10％ | 0.06751 | 0.42885 |
| 20％ | 0.06785 | 0.42916 |
| 30％ | 0.06829 | 0.42863 |
| 40％ | 0.06870 | 0.43487 |
| 50％ | 0.06871 | 0.43213 |
As shown in fig. 4 (a) to 4 (b), fig. 4 (a) shows a graph of the predicted result with the loss rate of 20%, fig. 4 (b) shows a trend and a periodicity, and from top to bottom are respectively: trend component, periodic component in units of weeks, periodic component in units of days.
As can be seen from table 2 and fig. 4 (a) to fig. 4 (b), the fitting effect of the optimized propset prediction model on the missing part is less affected by the missing rate, and the profset prediction model of the present invention has strong anti-interference capability on the missing value.
Example 2
(1) And collecting the original concentrations of dissolved hydrogen, methane, ethylene, acetylene, ethane, carbon monoxide, carbon dioxide and total hydrocarbon in a period from 8.8.8 to 6.8.6 of 2018 of transformer oil chromatographic online monitoring equipment, and obtaining 8713 pieces of oil chromatographic data containing time stamps.
(2) The judgment was made in the same manner as in step (2) in example 1: the above 8713 missing features of the oil chromatographic data containing the timestamp are completely associated with the missing of the multiple sequences, and belong to non-completely random missing, and the present embodiment still selects the hydrogen sequence to perform the trend prediction.
The data collected from 8.8.8.7.7.7.2018 is selected as training set data, and the data from 7.8.7.7.8.6.720 hours in 2019 is selected as test set data.
(3) And (3) establishing a Prophet prediction model according to the formulas (1) to (5).
(4) And (3) optimizing the Prophet prediction model established in the step (3), and finally selecting the group of parameters of lambda=0.1 and sigma=1 as optimal prediction model parameters.
(5) The hydrogen prediction result is obtained by taking the time sequence extension of the optimized Prophet prediction model in the step 4 as the output quantity, and the result is shown in fig. 5, and as can be seen from fig. 5, the prediction effect of the optimized Prophet prediction model for one month is still better, and the upper limit and the lower limit of the prediction effect can cover 93.75% of real data.
The invention has been described in detail with reference to preferred embodiments and illustrative examples. It should be noted, however, that these embodiments are merely illustrative of the present invention and do not limit the scope of the present invention in any way. Various improvements, equivalent substitutions or modifications can be made to the technical content of the present invention and its embodiments without departing from the spirit and scope of the present invention, which all fall within the scope of the present invention. The scope of the invention is defined by the appended claims.
Claims (10)
1. A method for predicting the concentration trend of dissolved gas in power transformer oil, the method comprising:
step 1, obtaining dissolved gas concentration model data in power transformer oil;
step 2, constructing training set data based on the dissolved gas concentration model data;
step 3, a propset prediction model is established;
step 4, optimizing the propet prediction model according to the training set data to obtain an optimized propet prediction model;
and 5, predicting the concentration trend of the dissolved gas according to the optimized Prophet prediction model.
2. The method according to claim 1, wherein in step 1, the concentration of the dissolved gas in the power transformer oil is subjected to an initial sequence of pretreatment and outlier cleaning to obtain model data of the concentration of the dissolved gas in the power transformer oil.
3. The method of claim 2, wherein the preprocessing comprises: acquiring the original concentration of dissolved gas in the power transformer oil, performing missing verification, and marking missing unknown data, non-conforming data format and non-conforming historical data as missing values; then, timing verification is performed.
4. The method of claim 2, wherein the outlier cleaning comprises: the outlier is converted into a missing value.
5. The method according to claim 1, wherein in step 2, a missing state analysis is performed based on the dissolved gas concentration model data; and judging the missing characteristics of the oil chromatographic data according to the missing state, and further constructing training set data.
6. The method of claim 5, wherein the missing state analysis comprises: the sequence occurrence missing value of the dissolved gas concentration model data is marked as 0, and the correlation of the time sequence of occurrence 0 is used as a missing thermodynamic diagram.
7. The method of claim 5, wherein the missing features of the oil chromatography data include a completely random loss and a non-completely random loss.
8. The method of claim 1, wherein in step 3, the propset predictive model is represented by trend, period, uncertainty, and error terms of dissolved gas concentration.
9. The method according to claim 1 or 8, characterized in that in step 4, the optimization of the Prophet prediction model is achieved by adjusting trend parameters and periodicity parameters.
10. The method according to claim 1, characterized in that in step 5, the optimum Prophet prediction model is used for carrying out time-series extension prediction to obtain the gas concentration to be predicted, so as to realize the prediction of the trend of the concentration of the dissolved gas in the power transformer oil.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311582300.0A CN117672403A (en) | 2023-11-24 | 2023-11-24 | Method for predicting concentration trend of dissolved gas in power transformer oil |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311582300.0A CN117672403A (en) | 2023-11-24 | 2023-11-24 | Method for predicting concentration trend of dissolved gas in power transformer oil |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117672403A true CN117672403A (en) | 2024-03-08 |
Family
ID=90081849
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311582300.0A Pending CN117672403A (en) | 2023-11-24 | 2023-11-24 | Method for predicting concentration trend of dissolved gas in power transformer oil |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117672403A (en) |
-
2023
- 2023-11-24 CN CN202311582300.0A patent/CN117672403A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20230015731A1 (en) | Real-time abnormity-diagnosing and interpolation method for water regime-monitoring data | |
| CN107730395B (en) | Power consumption abnormity detection method based on power consumption deviation rate for low-voltage users | |
| Garvey et al. | Validation of on-line monitoring techniques to nuclear plant data | |
| CN111191193A (en) | Long-term soil temperature and humidity high-precision prediction method based on autoregressive moving average model | |
| CN113672606A (en) | Oil chromatography monitoring data quality evaluation method | |
| CN111582551A (en) | Method and system for predicting short-term wind speed of wind power plant and electronic equipment | |
| CN113298297A (en) | Wind power output power prediction method based on isolated forest and WGAN network | |
| CN109100429A (en) | A kind of oil dissolved gas prediction technique of combined prediction residual GM | |
| CN112861350B (en) | Temperature overheating defect early warning method for stator winding of water-cooled steam turbine generator | |
| CN107067100A (en) | Wind power anomalous data identification method and device for identifying | |
| CN115587672A (en) | Distribution transformer load prediction and heavy overload early warning method and system | |
| CN114397526A (en) | Power transformer fault prediction method and system driven by state holographic sensing data | |
| CN111861206A (en) | Industrial industry prosperity index obtaining method based on enterprise electric power big data | |
| CN116432123A (en) | Electric energy meter fault early warning method based on CART decision tree algorithm | |
| CN113539382A (en) | Early warning positioning method and system for key process parameters of dimethyl phosphite | |
| CN115374938A (en) | XGboost-based power distribution network voltage prediction method | |
| CN112417627A (en) | Power distribution network operation reliability analysis method based on four-dimensional index system | |
| CN110188399B (en) | Dam safety monitoring single-measuring-point evaluation method based on multiple correlation sequences | |
| CN113866552B (en) | Medium voltage distribution network user electricity consumption abnormality diagnosis method based on machine learning | |
| CN111192163B (en) | Generator reliability medium-short term prediction method based on wind turbine generator operating data | |
| CN110196456A (en) | A kind of medium-term and long-term rainfall runoff forecasting method based on analog year grey correlation analysis | |
| CN117350897A (en) | Method and system for predicting conversion power by considering air humidity | |
| CN117200223A (en) | Day-ahead power load prediction method and device | |
| CN117672403A (en) | Method for predicting concentration trend of dissolved gas in power transformer oil | |
| CN115577854A (en) | Quantile regression wind speed interval prediction method based on EEMD-RBF combination |
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 |