CN110441500A - A kind of predicting model for dissolved gas in transformer oil method based on shot and long term memory network - Google Patents
A kind of predicting model for dissolved gas in transformer oil method based on shot and long term memory network Download PDFInfo
- Publication number
- CN110441500A CN110441500A CN201910751886.6A CN201910751886A CN110441500A CN 110441500 A CN110441500 A CN 110441500A CN 201910751886 A CN201910751886 A CN 201910751886A CN 110441500 A CN110441500 A CN 110441500A
- Authority
- CN
- China
- Prior art keywords
- shot
- long term
- term memory
- memory network
- 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.)
- Pending
Links
- 230000007787 long-term memory Effects 0.000 title claims abstract description 54
- 238000000034 method Methods 0.000 title claims abstract description 28
- 239000007789 gas Substances 0.000 claims abstract description 50
- 238000004587 chromatography analysis Methods 0.000 claims abstract description 8
- VNWKTOKETHGBQD-UHFFFAOYSA-N methane Chemical compound C VNWKTOKETHGBQD-UHFFFAOYSA-N 0.000 claims description 36
- CURLTUGMZLYLDI-UHFFFAOYSA-N Carbon dioxide Chemical compound O=C=O CURLTUGMZLYLDI-UHFFFAOYSA-N 0.000 claims description 34
- 238000012360 testing method Methods 0.000 claims description 32
- VGGSQFUCUMXWEO-UHFFFAOYSA-N Ethene Chemical compound C=C VGGSQFUCUMXWEO-UHFFFAOYSA-N 0.000 claims description 29
- HSFWRNGVRCDJHI-UHFFFAOYSA-N alpha-acetylene Natural products C#C HSFWRNGVRCDJHI-UHFFFAOYSA-N 0.000 claims description 27
- 238000012549 training Methods 0.000 claims description 23
- 239000005977 Ethylene Substances 0.000 claims description 17
- 238000011156 evaluation Methods 0.000 claims description 17
- 125000002534 ethynyl group Chemical group [H]C#C* 0.000 claims description 15
- 229910052739 hydrogen Inorganic materials 0.000 claims description 15
- 239000001257 hydrogen Substances 0.000 claims description 15
- UGFAIRIUMAVXCW-UHFFFAOYSA-N Carbon monoxide Chemical compound [O+]#[C-] UGFAIRIUMAVXCW-UHFFFAOYSA-N 0.000 claims description 13
- 229910002092 carbon dioxide Inorganic materials 0.000 claims description 13
- 239000001569 carbon dioxide Substances 0.000 claims description 13
- 229910002091 carbon monoxide Inorganic materials 0.000 claims description 13
- OTMSDBZUPAUEDD-UHFFFAOYSA-N Ethane Chemical compound CC OTMSDBZUPAUEDD-UHFFFAOYSA-N 0.000 claims description 12
- 150000002431 hydrogen Chemical class 0.000 claims description 12
- 210000002569 neuron Anatomy 0.000 claims description 9
- 238000005457 optimization Methods 0.000 claims description 8
- 238000006243 chemical reaction Methods 0.000 claims description 6
- 230000006870 function Effects 0.000 claims description 5
- 238000004458 analytical method Methods 0.000 claims description 4
- 238000010606 normalization Methods 0.000 claims description 4
- 238000012545 processing Methods 0.000 claims description 4
- UFHFLCQGNIYNRP-UHFFFAOYSA-N Hydrogen Chemical compound [H][H] UFHFLCQGNIYNRP-UHFFFAOYSA-N 0.000 claims description 3
- 230000004913 activation Effects 0.000 claims description 3
- 238000002474 experimental method Methods 0.000 claims description 3
- 230000008569 process Effects 0.000 claims description 3
- 230000007774 longterm Effects 0.000 claims description 2
- 150000001336 alkenes Chemical class 0.000 claims 1
- 238000012546 transfer Methods 0.000 claims 1
- 238000012423 maintenance Methods 0.000 abstract description 5
- OKTJSMMVPCPJKN-UHFFFAOYSA-N Carbon Chemical compound [C] OKTJSMMVPCPJKN-UHFFFAOYSA-N 0.000 description 11
- 229910052799 carbon Inorganic materials 0.000 description 11
- 229960004424 carbon dioxide Drugs 0.000 description 9
- 229910002090 carbon oxide Inorganic materials 0.000 description 9
- 238000013528 artificial neural network Methods 0.000 description 6
- 238000012706 support-vector machine Methods 0.000 description 4
- 239000004215 Carbon black (E152) Substances 0.000 description 3
- 230000032683 aging Effects 0.000 description 3
- 230000008901 benefit Effects 0.000 description 3
- 238000010292 electrical insulation Methods 0.000 description 3
- 229930195733 hydrocarbon Natural products 0.000 description 3
- 150000002430 hydrocarbons Chemical class 0.000 description 3
- 239000007787 solid Substances 0.000 description 3
- 230000000052 comparative effect Effects 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 241001269238 Data Species 0.000 description 1
- PEDCQBHIVMGVHV-UHFFFAOYSA-N Glycerine Chemical compound OCC(O)CO PEDCQBHIVMGVHV-UHFFFAOYSA-N 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000004090 dissolution Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000004868 gas analysis Methods 0.000 description 1
- 230000015654 memory Effects 0.000 description 1
- 230000003647 oxidation Effects 0.000 description 1
- 238000007254 oxidation reaction Methods 0.000 description 1
- 230000008439 repair process Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N33/00—Investigating or analysing materials by specific methods not covered by groups G01N1/00 - G01N31/00
- G01N33/26—Oils; Viscous liquids; Paints; Inks
- G01N33/28—Oils, i.e. hydrocarbon liquids
- G01N33/2835—Specific substances contained in the oils or fuels
- G01N33/2841—Gas in oils, e.g. hydrogen in insulating oils
Landscapes
- Chemical & Material Sciences (AREA)
- Health & Medical Sciences (AREA)
- Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Medicinal Chemistry (AREA)
- Oil, Petroleum & Natural Gas (AREA)
- General Chemical & Material Sciences (AREA)
- Food Science & Technology (AREA)
- Chemical Kinetics & Catalysis (AREA)
- Physics & Mathematics (AREA)
- Analytical Chemistry (AREA)
- Biochemistry (AREA)
- General Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of predicting model for dissolved gas in transformer oil methods based on shot and long term memory network, by the concentration of Accurate Prediction Gases Dissolved in Transformer Oil, to realize the assessment to running state of transformer;The online oil chromatography sample data of transformer is collected first, is determined its characteristic parameter and is normalized;Then shot and long term memory network is trained using the sequence data of Gases Dissolved in Transformer Oil, obtains optimal prediction model parameters;It is finally input with the 7 kinds of characteristic gas concentration dissolved in oil, is output with the concentration of gas to be predicted, realizes the prediction to gas dissolved in oil of power trans-formers.Method provided by the invention can be provided foundation to the operation conditions judgement of power transformer, be offered reference for operation maintenance personnel maintenance with the variation of Accurate Prediction oil dissolved gas concentration.
Description
Technical field
The present invention relates to electric apparatus monitoring technical field more particularly to a kind of predicting model for dissolved gas in transformer oil
Method.
Background technique
Power transformer is as the equipment of core, the important function for playing distribution, transmitting electric energy are the most in electric system
The critical asset of grid company, safe and stable operation are the premises for ensureing electric network reliability power supply.Transformer operates normally feelings
Under condition, it is dissolved in insulating oil since the aging of built-in electrical insulation solid can generate a small amount of gas, mainly there is hydrogen (H2), methane
(CH4), ethane (C2H6), ethylene (C2H4), acetylene (C2H2), carbon monoxide (CO), carbon dioxide (CO2) etc. gases.According in oil
The difference of concentration of dissolved gas ratio can differentiate the different operation conditions of transformer, such as: hydrogen H when generating high-energy discharge2With
Acetylene C2H2Content will increase;In the case where encountering strong magnetic field, the content of hydrocarbon gas be will increase, and be shown certain
Association.
Being analyzed (dissolved gas analysis, DGA) to gas dissolved in oil of power trans-formers can transport for it
Row situation differentiates that important role, technique are proved by a large amount of practical diagnosis cases, is current internationally recognized one
Kind early stage transformer fault method of discrimination.Due to traditional back propagation artificial neural network model (Back Propagation
Neural Network, BPNN) and support vector machines (support vector machine, SVM) there are the periods for prediction technique
Compared with the disadvantages of long, error is big, convergence rate is slow, network structure selection is different, cannot forecast analysis dissolved gas in time concentration,
Therefore delay judgement can be generated to the operation conditions of transformer, lead to economic loss.
Summary of the invention
The technical problem to be solved in the present invention is to provide solution gas in a kind of transformer oil based on shot and long term memory network
Bulk concentration prediction technique, the parameter for needing to adjust using the prediction technique is few, while can handle the non-linear sequence of long period
Column forecasting problem, can promptly and accurately prediction Gases Dissolved in Transformer Oil concentration, thus realize to transformer station high-voltage side bus shape
The assessment of condition.
In order to solve the above technical problems, the present invention provides dissolve in a kind of transformer oil based on shot and long term memory network
Forecasting of Gas Concentration method, includes the following steps:
A, the hydrogen (H dissolved in transformer oil is selected2) concentration, methane (CH4) concentration, ethane (C2H6) concentration, ethylene
(C2H4) concentration, acetylene (C2H2) concentration, carbon monoxide (CO) concentration, carbon dioxide (CO2) concentration is as characteristic parameter;
B, chromatography sample data of the characteristic parameter in obtaining step A in historical time dimension transformer oil, and to sample
Notebook data is normalized, and divides training set data and test set data;
C, building includes the shot and long term memory network prediction model LSTM of input layer, hidden layer, output layer;
D, using the training set data obtained in step B to the shot and long term memory network prediction model constructed in step C
LSTM is optimized, and determines the parameter of shot and long term memory network prediction model LSTM;
E, use test set data in step B as shot and long term memory network prediction model optimized in step D
The input variable of LSTM, one of characteristic parameter in selecting step A is as shot and long term memory network prediction optimized in step D
The output variable of model LSTM, obtains prediction result.
Finger is normalized to sample data in the step B sample data is mapped between [0,1], converts letter
Number are as follows:
X in formulaminFor sample data minimum value, xmaxFor sample data maximum value, x is the sample data before conversion, x*For
Sample data after conversion;
Training set data is divided in the step B and test set data refer to 80% of the sample data after normalized
As training set data, test set data are used as by 20% of the sample data after normalized.
The specific building process of shot and long term memory network prediction model LSTM is as follows in the step C:
C1, the online oil chromatography sample data of transformer is collected, takes characteristic parameter hydrogen (H2), methane (CH4), ethane
(C2H6), ethylene (C2H4), acetylene (C2H2), carbon monoxide (CO), carbon dioxide (CO2) input layer of the concentration as model
Sample data;
C2, the sample data of each characteristic parameter is normalized, is mapped between [0,1], divide training set number
According to test set data;
C3, hidden layer instruct shot and long term memory network prediction model LSTM using training set data and test set data
Practice, test, the shot and long term memory network prediction model in the lesser situation of prediction result evaluation criterion error, after determining optimization
The parameter of LSTM;
Shot and long term memory network prediction model LSTM after C4, the optimization obtained using step C3 carries out test data set
Prediction, obtains prediction result P1、P2、…Pn;
C5, output layer are used and are calculated n prediction result in a manner of average value, and are carried out anti-normalization processing and obtained finally
Prediction result, last Binding experiment evaluation index to prediction result carry out evaluation analysis.
The shot and long term memory network prediction model LSTM of building is optimized in the step D, it is average opposite in guarantee
Percent error yMAPE, root mean square error yRMSEMinimum, precision of prediction yFAUnder the premise of highest, determine that shot and long term memory network is predicted
The optimized parameter of model LSTM;Specific steps include: using ReLU function as activation primitive in hidden layer, be suitable for time sequence
Column nonlinear prediction, initial learning rate are set as 0.001, and the rejection rate of each layer network node is 0.2, prevent overfitting,
The number of iterations takes 200, and the number of neuron is determined by the feature of training set data between input layer and output layer, using experience public affairs
The number that neuron is chosen under conditions of formula (2) and evaluation index are optimal is 10, and then controlling neuron number is 10 constant
In the case of, it is stepped up the network number of plies and carrys out test model, finally combine the evaluation index of average relative error, choosing network layer is
2;
In formula, n and m are respectively the number of nodes of output layer and input layer, constant of a between [0,10].
Prediction result obtained in the step E uses average percentage error yMAPE, root mean square error yRMSEAnd it is pre-
Survey precision yFAThree experimental evaluation indexs are evaluated, and formula is as follows:
In formula: n indicates prediction total degree;Xact(i) and XpredIt (i) is respectively the true of i moment oil dissolved gas concentration
Value and predicted value, wherein averagely percentage error and root mean square error are smaller, the higher expression prediction result of precision of prediction is better.
Prediction technique provided by the invention has following advantage compared with prior art: proposed by the present invention to be based on shot and long term
The predicting model for dissolved gas in transformer oil method of memory network, the variation for capableing of Accurate Prediction oil dissolved gas concentration become
Gesture provides foundation to the operation conditions judgement of power transformer, can reduce the probability that transformer breaks down, examine for operation maintenance personnel
It repairs and offers reference.
Detailed description of the invention
Fig. 1 is the model schematic of the predicting model for dissolved gas in transformer oil method based on shot and long term memory network;
Fig. 2 is the prediction curve and reality for predicting ethylene concentration in embodiment using shot and long term memory network prediction model LSTM
Border curve comparison figure;
Shot and long term memory network prediction model LSTM, BPNN neural network, support vector machine is used in the embodiment of the position Fig. 3
SVM predicts the prediction curve and actual curve comparison diagram of ethylene concentration respectively.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is
A part of the embodiments of the present invention, instead of all the embodiments;Based on the embodiments of the present invention, ordinary skill people
Member's every other embodiment obtained without creative efforts, shall fall within the protection scope of the present invention.
The present invention provides a kind of predicting model for dissolved gas in transformer oil methods of shot and long term memory network, including such as
Lower step:
A, the hydrogen (H dissolved in transformer oil is selected2) concentration, methane (CH4) concentration, ethane (C2H6) concentration, ethylene
(C2H4) concentration, acetylene (C2H2) concentration, carbon monoxide (CO) concentration, carbon dioxide (CO2) concentration is as characteristic parameter;
Under power transformer normal operation, it is dissolved in absolutely since the aging of built-in electrical insulation solid can generate a small amount of gas
In edge oil, mainly there is hydrogen (H2) concentration, methane (CH4) concentration, ethane (C2H6) concentration, ethylene (C2H4) concentration, acetylene (C2H2)
Concentration, carbon monoxide (CO) concentration, carbon dioxide (CO2) etc. gases.It can be sentenced according to the difference of Oil Dissolved Gases Concentration ratio
The different operation conditions of other transformer, such as: hydrogen H when generating high-energy discharge2With acetylene C2H2Content will increase;It encounters strong
In the case where strong magnetic field, the content of hydrocarbon gas be will increase, and show certain association, and the present invention selects hydrogen (H2) dense
Degree, methane (CH4) concentration, ethane (C2H6) concentration, ethylene (C2H4) concentration, acetylene (C2H2) concentration, carbon monoxide (CO) concentration,
Carbon dioxide (CO2) etc. 7 kinds of dissolved gas as characteristic parameter.
B, chromatography sample data of the characteristic parameter in obtaining step A in historical time dimension transformer oil, and to sample
Notebook data is normalized, and divides training set data and test set data;
Finger is normalized to sample data in the step B sample data is mapped between [0,1], converts letter
Number are as follows:
X in formulaminFor sample data minimum value, xmaxFor sample data maximum value, x is the sample data before conversion, x*For
Sample data after conversion;
Training set data is divided in the step B and test set data refer to 80% of the sample data after normalized
As training set data, test set data are used as by 20% of the sample data after normalized.
The concentration of 7 kinds of dissolved gas involved in the present invention, sample data have 250, and every number of cases evidence has 7 characteristic parameters
(hydrogen (H2) concentration, methane (CH4) concentration, ethane (C2H6) concentration, ethylene (C2H4) concentration, acetylene (C2H2) concentration, an oxidation
Carbon (CO) concentration, carbon dioxide (CO2) concentration), using 80% (200) of sample data set as training set data, 20%
(50) are used as test set data, and the shot and long term memory network prediction model of building is trained and is tested respectively.
C, building includes the shot and long term memory network prediction model LSTM of input layer, hidden layer, output layer;
The specific building process of shot and long term memory network prediction model LSTM is as follows in the step C:
C1, the online oil chromatography sample data of transformer is collected, takes characteristic parameter hydrogen (H2), methane (CH4), ethane
(C2H6), ethylene (C2H4), acetylene (C2H2), carbon monoxide (CO), carbon dioxide (CO2) input layer of the concentration as model
Sample data;
C2, the sample data of each characteristic parameter is normalized, is mapped between [0,1], divide training set number
According to test set data;
C3, hidden layer instruct shot and long term memory network prediction model LSTM using training set data and test set data
Practice, test, the shot and long term memory network prediction model in the lesser situation of prediction result evaluation criterion error, after determining optimization
The parameter of LSTM;
Shot and long term memory network prediction model LSTM after C4, the optimization obtained using step C3 carries out test data set
Prediction, obtains prediction result P1、P2、…Pn;
C5, output layer are used and are calculated n prediction result in a manner of average value, and are carried out anti-normalization processing and obtained finally
Prediction result, last Binding experiment evaluation index to prediction result carry out evaluation analysis.
D, using the training set data obtained in step B to the shot and long term memory network prediction model constructed in step C
LSTM is optimized, and determines the parameter of shot and long term memory network prediction model LSTM;
The shot and long term memory network prediction model LSTM of building is optimized in the step D, it is average opposite in guarantee
Percent error yMAPE, root mean square error yRMSEMinimum, precision of prediction yFAUnder the premise of highest, determine that shot and long term memory network is predicted
The optimized parameter of model LSTM;Specific steps include: using ReLU function as activation primitive in hidden layer, be suitable for time sequence
Column nonlinear prediction, initial learning rate are set as 0.001, and the rejection rate of each layer network node is 0.2, prevent overfitting,
The number of iterations takes 200, and the number of neuron is determined by the feature of training set data between input layer and output layer, using experience public affairs
The number that neuron is chosen under conditions of formula (2) and evaluation index are optimal is 10, and then controlling neuron number is 10 constant
In the case of, it is stepped up the network number of plies and carrys out test model, finally combine the evaluation index of average relative error, choosing network layer is
2;
In formula, n and m are respectively the number of nodes of output layer and input layer, constant of a between [0,10].
E, use test set data in step B as shot and long term memory network prediction model optimized in step D
The input variable of LSTM, one of characteristic parameter in selecting step A is as shot and long term memory network prediction optimized in step D
The output variable of model LSTM, obtains prediction result.
Test set data are predicted using the shot and long term memory network model after optimization, obtain prediction result P1、
P2、…Pn, use and calculate n prediction result in a manner of average value, and carry out anti-normalization processing and obtain final prediction knot
Fruit.
Obtained final prediction result uses average percentage error yMAPE, root mean square error yRMSEAnd precision of prediction
yFAThree indexs are evaluated, and formula is as follows:
In formula: n indicates prediction total degree;Xact(i) and XpredIt (i) is respectively the true of i moment oil dissolved gas concentration
Value and predicted value, wherein averagely percentage error and root mean square error are smaller, the higher expression prediction result of precision of prediction is better.
Embodiment
Under power transformer normal operation, it is dissolved in absolutely since the aging of built-in electrical insulation solid can generate a small amount of gas
In edge oil, mainly there is hydrogen (H2), methane (CH4), ethane (C2H6), ethylene (C2H4), acetylene (C2H2), carbon monoxide (CO), two
Carbonoxide (CO2) etc. gases.The different operation conditions of transformer, example can be differentiated according to the difference of Oil Dissolved Gases Concentration ratio
Such as: hydrogen H when generating high-energy discharge2With acetylene C2H2Content will increase;In the case where encountering strong magnetic field, hydrocarbon gas
Content will increase, and show certain association, and the present invention selects hydrogen (H2), methane (CH4), ethane (C2H6), ethylene
(C2H4), acetylene (C2H2), carbon monoxide (CO), carbon dioxide (CO2) etc. 7 kinds of gases as characteristic parameter.
The present embodiment is directed to the prediction of gas dissolved in oil of power trans-formers, and collection characteristic parameter is tieed up in historical time first
The chromatography sample data in transformer oil is spent, characteristic parameter is determined and is normalized;Then using molten in transformer oil
The sample data of solution characteristic parameter is trained shot and long term memory network model, obtains optimal prediction model parameters;Finally
Using the concentration of the 7 kinds of characteristic parameters dissolved in transformer oil as input variable, using the concentration of one of characteristic parameter as output variable,
Realize the prediction of gas dissolved in oil of power trans-formers.It is predicted under different models using above-mentioned sample data, verifying is originally
Invent the validity of the predicting model for dissolved gas in transformer oil method based on shot and long term memory network proposed.
By taking certain 220KV transformer oil chromatographic online monitoring data as an example, sample data is on August 1, to 2018 4 2017
The moon 10, amount to 250 groups of detection datas, detection cycle is 1 day;By totally 200 groups of prisons in August in 2017 on January 20th, 1 day 1
Measured data as training set data, using on April 10,21 days to 2018 January in 2018 totally 50 groups of data as test set data.
Fig. 3 is to algorithms of different model using 50 groups of test sets to dissolution ethylene C in transformer oil2H4The predicted value of concentration
With the line chart of true value comparison.It can be seen from the figure that related 3 kinds of algorithms shot and long term memory network LSTM, support to
Amount machine (support vector machine, SVM), back propagation artificial neural network model (Back Propagation Neural
Network, BPNN) there is preferable estimated performance, what the shot and long term memory network LSTM prediction model that the present invention is mentioned obtained
Ethylene C2H4Concentration prediction curve precision highest, it is almost the same with real gas changes of contents trend.
Combining assessment index, the LSTM Network Prediction Model and SVM, BPNN Network Prediction Model that the present invention is mentioned are predicted
Ethylene C2H4The Comparative result of concentration is as shown in table 1, and the average relative error that can obtain LSTM prediction is 0.705%, and precision of prediction is
99.295, no matter equal in terms of average relative error, root mean square error and precision of prediction it is based on LSTM network model prediction technique
Better than other 2 kinds of methods, it is known that the LSTM network model that the present invention is mentioned is more acurrate on precision of prediction, and prediction effect is more preferable.
The different model prediction Comparative results of table 1
Model | Average relative error (%) | Root mean square error | Precision of prediction |
LSTM | 0.705 | 0.117 | 99.295 |
SVM | 3.511 | 0.439 | 96.489 |
BPNN | 4.165 | 0.497 | 95.835 |
Use LSTM prediction model with the data instance between on the April 10th, 1 day 1 of August in 2017, it is several to other
Kind gas concentration is predicted that the results are shown in Table 2: knowing that the average relative error for the LSTM prediction model that the present invention is mentioned is equal
Lower than SVM, BPNN prediction model, precision of prediction is highest in 3 kinds of models, predicting reliability with higher.
Other forecasting of Gas Concentration results of table 2
Table 3 is test sample collection data continuous 20 days under LSTM network model, SVM network model and BPNN network model
Ethylene concentration actual value and predicted value and absolute percent error summary sheet, we can see that LSTM network mould from table
The mean error minimum 0.600 of type prediction, the mean error of respectively less than SVM and BPNN prediction model, it is known that the present invention is mentioned
LSTM model prediction resultant error it is minimum, be best suitable for actual value.
3 actual value of table and predicted value and absolute percent are missed
Note: true value, the unit of predicted value are equal in upper table are as follows: (μ L/L) is proposed by the present invention a kind of based on shot and long term memory
The predicting model for dissolved gas in transformer oil method of network is inherited with the concentration of Accurate Prediction Gases Dissolved in Transformer Oil
The advantages that traditional neural network prediction model is easy to operate, and stability is good, and prediction error is small, to the non-linear sequence of long period
Column prediction has higher accuracy rate.By analyzing gas prediction result, can judge to provide for the operation conditions of power transformer
Foundation is offered reference for operation maintenance personnel maintenance.
The above is only a preferred embodiment of the present invention, it should be pointed out that: for the ordinary skill people of the art
For member, various improvements and modifications may be made without departing from the principle of the present invention, these improvements and modifications are also answered
It is considered as protection scope of the present invention.
Claims (5)
1. a kind of predicting model for dissolved gas in transformer oil method based on shot and long term memory network, which is characterized in that including
Following steps:
A, the hydrogen (H dissolved in transformer oil is selected2) concentration, methane (CH4) concentration, ethane (C2H6) concentration, ethylene (C2H4)
Concentration, acetylene (C2H2) concentration, carbon monoxide (CO) concentration, carbon dioxide (CO2) concentration is as characteristic parameter;
B, chromatography sample data of the characteristic parameter in obtaining step A in historical time dimension transformer oil, and to sample number
According to being normalized, training set data and test set data are divided;
C, building includes the shot and long term memory network prediction model LSTM of input layer, hidden layer, output layer;
D, using the training set data obtained in step B to the shot and long term memory network prediction model LSTM constructed in step C into
Row optimization, determines the parameter of shot and long term memory network prediction model LSTM;
E, use the test set data in step B as shot and long term memory network prediction model LSTM's optimized in step D
Input variable, one of characteristic parameter in selecting step A is as shot and long term memory network prediction model optimized in step D
The output variable of LSTM, obtains prediction result.
2. a kind of predicting model for dissolved gas in transformer oil side based on shot and long term memory network according to claim 1
Method, it is characterised in that: finger is normalized to sample data in the step B, sample data is mapped between [0,1],
Transfer function are as follows:
X in formulaminFor sample data minimum value, xmaxFor sample data maximum value, x is the sample data before conversion, x*After conversion
Sample data;
Training set data is divided in the step B and test set data refer to 80% conduct of the sample data after normalized
Training set data is used as test set data for 20% of the sample data after normalized.
3. a kind of predicting model for dissolved gas in transformer oil side based on shot and long term memory network according to claim 1
Method, it is characterised in that: the specific building process of shot and long term memory network prediction model LSTM is as follows in the step C:
C1, the online oil chromatography sample data of transformer is collected, takes characteristic parameter hydrogen (H2), methane (CH4), ethane (C2H6), second
Alkene (C2H4), acetylene (C2H2), carbon monoxide (CO), carbon dioxide (CO2) concentration as model input layer sample data;
C2, the sample data of each characteristic parameter is normalized, is mapped between [0,1], divide training set data and
Test set data;
C3, hidden layer be trained shot and long term memory network prediction model LSTM using training set data and test set data,
Test, the shot and long term memory network prediction model in the lesser situation of prediction result evaluation criterion error, after determining optimization
The parameter of LSTM;
Shot and long term memory network prediction model LSTM after C4, the optimization obtained using step C3 carries out test data set pre-
It surveys, obtains prediction result P1、P2、…Pn;
C5, output layer are used and are calculated n prediction result in a manner of average value, and carry out anti-normalization processing obtain it is final pre-
It surveys as a result, last Binding experiment evaluation index carries out evaluation analysis to prediction result.
4. a kind of predicting model for dissolved gas in transformer oil side based on shot and long term memory network according to claim 1
Method, it is characterised in that: the shot and long term memory network prediction model LSTM of building is optimized in the step D, is guaranteeing to put down
With respect to percent error yMAPE, root mean square error yRMSEMinimum, precision of prediction yFAUnder the premise of highest, determine that shot and long term remembers net
The optimized parameter of network prediction model LSTM;Specific steps include: using ReLU function as activation primitive in hidden layer, be suitable for
Time series nonlinear prediction, initial learning rate are set as 0.001, and the rejection rate of each layer network node is 0.2, is prevented excessively
Fitting, the number of iterations take 200, and the number of neuron is determined by the feature of training set data between input layer and output layer, use
The number that neuron is chosen under conditions of empirical equation (2) and evaluation index are optimal is 10, and then controlling neuron number is 10
It in the case where constant, be stepped up the network number of plies and carry out test model, finally combine the evaluation index of average relative error, choose net
Network layers are 2;
In formula, n and m are respectively the number of nodes of output layer and input layer, constant of a between [0,10].
5. a kind of predicting model for dissolved gas in transformer oil side based on shot and long term memory network according to claim 1
Method, it is characterised in that: prediction result obtained in the step E uses average percentage error yMAPE, root mean square error yRMSE
And precision of prediction yFAThree experimental evaluation indexs are evaluated, and formula is as follows:
In formula: n indicates prediction total degree;Xact(i) and Xpred(i) be respectively i moment oil dissolved gas concentration true value and
Predicted value, wherein averagely percentage error and root mean square error are smaller, the higher expression prediction result of precision of prediction is better.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910751886.6A CN110441500A (en) | 2019-08-15 | 2019-08-15 | A kind of predicting model for dissolved gas in transformer oil method based on shot and long term memory network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910751886.6A CN110441500A (en) | 2019-08-15 | 2019-08-15 | A kind of predicting model for dissolved gas in transformer oil method based on shot and long term memory network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110441500A true CN110441500A (en) | 2019-11-12 |
Family
ID=68435638
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910751886.6A Pending CN110441500A (en) | 2019-08-15 | 2019-08-15 | A kind of predicting model for dissolved gas in transformer oil method based on shot and long term memory network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110441500A (en) |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111337768A (en) * | 2020-03-02 | 2020-06-26 | 武汉大学 | Deep parallel fault diagnosis method and system for dissolved gas in transformer oil |
CN111426816A (en) * | 2020-04-10 | 2020-07-17 | 昆明理工大学 | Method for predicting concentration of dissolved gas in transformer oil based on PSO-L STM |
CN111985707A (en) * | 2020-08-17 | 2020-11-24 | 云南电力技术有限责任公司 | Method and device for predicting gas concentration in insulating oil |
CN112329335A (en) * | 2020-10-22 | 2021-02-05 | 杭州电力设备制造有限公司 | Long-term prediction method for content of dissolved gas in transformer oil |
CN112734028A (en) * | 2020-12-28 | 2021-04-30 | 三峡大学 | Modeling method for prediction model of concentration of dissolved gas in transformer oil |
CN112946187A (en) * | 2021-01-22 | 2021-06-11 | 西安科技大学 | Refuge chamber real-time state monitoring method based on neural network |
CN113673766A (en) * | 2021-08-23 | 2021-11-19 | 国网山西省电力公司晋城供电公司 | Method for predicting gas content in oil of oil-filled electrical equipment |
CN113780420A (en) * | 2021-09-10 | 2021-12-10 | 湖南大学 | Method for predicting concentration of dissolved gas in transformer oil based on GRU-GCN |
CN113849540A (en) * | 2021-09-22 | 2021-12-28 | 广东电网有限责任公司 | Fault prediction model training and prediction method, device, electronic equipment and medium |
CN113948159A (en) * | 2021-12-21 | 2022-01-18 | 云智慧(北京)科技有限公司 | Fault detection method, device and equipment for transformer |
CN116204794A (en) * | 2023-05-04 | 2023-06-02 | 国网江西省电力有限公司电力科学研究院 | Method and system for predicting dissolved gas in transformer oil by considering multidimensional data |
CN117408166A (en) * | 2023-12-14 | 2024-01-16 | 沧州经济开发区武理工京津冀协同产业科技研究院 | Method for predicting concentration of dissolved gas in transformer oil |
Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101949892A (en) * | 2010-08-20 | 2011-01-19 | 中国人民解放军第三军医大学第三附属医院 | Gas concentration detection method based on RBF (Radial Basis Function) artificial neural network and SAW (Surface Acoustic Wave) gas sensor |
CN102778555A (en) * | 2012-08-06 | 2012-11-14 | 国网技术学院 | Method for predicting concentration of gas dissolved in transformer oil |
CN105241497A (en) * | 2015-09-23 | 2016-01-13 | 国网山东省电力公司日照供电公司 | Transformer monitoring system and fault diagnosis method |
CN105259435A (en) * | 2015-09-23 | 2016-01-20 | 国网山东莒县供电公司 | Transformer monitoring device and fault diagnosis method |
CN106153567A (en) * | 2016-06-22 | 2016-11-23 | 南京信息工程大学 | Based on the pressure compensated infrared gas sensor of BP neutral net and detection method |
CN107545307A (en) * | 2017-07-28 | 2018-01-05 | 上海交通大学 | Predicting model for dissolved gas in transformer oil method and system based on depth belief network |
CN105930901B (en) * | 2016-07-18 | 2018-05-25 | 河海大学 | A kind of Diagnosis Method of Transformer Faults based on RBPNN |
CN105353255B (en) * | 2015-11-27 | 2018-07-31 | 南京邮电大学 | A kind of Diagnosis Method of Transformer Faults based on neural network |
CN108426812A (en) * | 2018-04-08 | 2018-08-21 | 浙江工业大学 | A kind of PM2.5 concentration value prediction techniques based on Memory Neural Networks |
CN106548230B (en) * | 2016-10-14 | 2019-08-06 | 云南电网有限责任公司昆明供电局 | Diagnosis Method of Transformer Faults based on Modified particle swarm optimization neural network |
-
2019
- 2019-08-15 CN CN201910751886.6A patent/CN110441500A/en active Pending
Patent Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101949892A (en) * | 2010-08-20 | 2011-01-19 | 中国人民解放军第三军医大学第三附属医院 | Gas concentration detection method based on RBF (Radial Basis Function) artificial neural network and SAW (Surface Acoustic Wave) gas sensor |
CN102778555A (en) * | 2012-08-06 | 2012-11-14 | 国网技术学院 | Method for predicting concentration of gas dissolved in transformer oil |
CN105241497A (en) * | 2015-09-23 | 2016-01-13 | 国网山东省电力公司日照供电公司 | Transformer monitoring system and fault diagnosis method |
CN105259435A (en) * | 2015-09-23 | 2016-01-20 | 国网山东莒县供电公司 | Transformer monitoring device and fault diagnosis method |
CN105353255B (en) * | 2015-11-27 | 2018-07-31 | 南京邮电大学 | A kind of Diagnosis Method of Transformer Faults based on neural network |
CN106153567A (en) * | 2016-06-22 | 2016-11-23 | 南京信息工程大学 | Based on the pressure compensated infrared gas sensor of BP neutral net and detection method |
CN105930901B (en) * | 2016-07-18 | 2018-05-25 | 河海大学 | A kind of Diagnosis Method of Transformer Faults based on RBPNN |
CN106548230B (en) * | 2016-10-14 | 2019-08-06 | 云南电网有限责任公司昆明供电局 | Diagnosis Method of Transformer Faults based on Modified particle swarm optimization neural network |
CN107545307A (en) * | 2017-07-28 | 2018-01-05 | 上海交通大学 | Predicting model for dissolved gas in transformer oil method and system based on depth belief network |
CN108426812A (en) * | 2018-04-08 | 2018-08-21 | 浙江工业大学 | A kind of PM2.5 concentration value prediction techniques based on Memory Neural Networks |
Non-Patent Citations (2)
Title |
---|
代杰杰 等: "采用LSTM 网络的电力变压器运行状态预测方法研究", 《高电压技术》 * |
刘云鹏 等: "基于经验模态分解和长短期记忆神经网络的变压器油中溶解气体浓度预测方法", 《中国电机工程学报》 * |
Cited By (20)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111337768A (en) * | 2020-03-02 | 2020-06-26 | 武汉大学 | Deep parallel fault diagnosis method and system for dissolved gas in transformer oil |
CN111337768B (en) * | 2020-03-02 | 2021-01-19 | 武汉大学 | Deep parallel fault diagnosis method and system for dissolved gas in transformer oil |
US11656298B2 (en) * | 2020-03-02 | 2023-05-23 | Wuhan University | Deep parallel fault diagnosis method and system for dissolved gas in transformer oil |
US20210278478A1 (en) * | 2020-03-02 | 2021-09-09 | Wuhan University | Deep parallel fault diagnosis method and system for dissolved gas in transformer oil |
CN111426816A (en) * | 2020-04-10 | 2020-07-17 | 昆明理工大学 | Method for predicting concentration of dissolved gas in transformer oil based on PSO-L STM |
CN111985707A (en) * | 2020-08-17 | 2020-11-24 | 云南电力技术有限责任公司 | Method and device for predicting gas concentration in insulating oil |
CN112329335A (en) * | 2020-10-22 | 2021-02-05 | 杭州电力设备制造有限公司 | Long-term prediction method for content of dissolved gas in transformer oil |
CN112329335B (en) * | 2020-10-22 | 2024-03-01 | 杭州电力设备制造有限公司 | Long-term prediction method for content of dissolved gas in transformer oil |
CN112734028A (en) * | 2020-12-28 | 2021-04-30 | 三峡大学 | Modeling method for prediction model of concentration of dissolved gas in transformer oil |
CN112946187A (en) * | 2021-01-22 | 2021-06-11 | 西安科技大学 | Refuge chamber real-time state monitoring method based on neural network |
CN112946187B (en) * | 2021-01-22 | 2023-04-07 | 西安科技大学 | Refuge chamber real-time state monitoring method based on neural network |
CN113673766A (en) * | 2021-08-23 | 2021-11-19 | 国网山西省电力公司晋城供电公司 | Method for predicting gas content in oil of oil-filled electrical equipment |
CN113673766B (en) * | 2021-08-23 | 2023-12-19 | 国网山西省电力公司晋城供电公司 | Method for predicting gas content in oil of oil-filled electrical equipment |
CN113780420A (en) * | 2021-09-10 | 2021-12-10 | 湖南大学 | Method for predicting concentration of dissolved gas in transformer oil based on GRU-GCN |
CN113780420B (en) * | 2021-09-10 | 2023-11-07 | 湖南大学 | GRU-GCN-based method for predicting concentration of dissolved gas in transformer oil |
CN113849540A (en) * | 2021-09-22 | 2021-12-28 | 广东电网有限责任公司 | Fault prediction model training and prediction method, device, electronic equipment and medium |
CN113948159A (en) * | 2021-12-21 | 2022-01-18 | 云智慧(北京)科技有限公司 | Fault detection method, device and equipment for transformer |
CN116204794B (en) * | 2023-05-04 | 2023-09-12 | 国网江西省电力有限公司电力科学研究院 | Method and system for predicting dissolved gas in transformer oil by considering multidimensional data |
CN116204794A (en) * | 2023-05-04 | 2023-06-02 | 国网江西省电力有限公司电力科学研究院 | Method and system for predicting dissolved gas in transformer oil by considering multidimensional data |
CN117408166A (en) * | 2023-12-14 | 2024-01-16 | 沧州经济开发区武理工京津冀协同产业科技研究院 | Method for predicting concentration of dissolved gas in transformer oil |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110441500A (en) | A kind of predicting model for dissolved gas in transformer oil method based on shot and long term memory network | |
CN109543737B (en) | Information system health degree evaluation method based on FAHP-FCA combined empowerment | |
CN106251059B (en) | Cable state evaluation method based on probabilistic neural network algorithm | |
CN102928720B (en) | Defect rate detecting method of oil immersed type main transformer | |
CN102809718A (en) | Ultra-high-frequency partial discharge signal identification method for gas insulated switchgear (GIS) | |
CN108876163B (en) | Transient state power angle stability rapid evaluation method integrating causal analysis and machine learning | |
CN110765268B (en) | Client appeal-based accurate distribution network investment strategy method | |
CN107274067B (en) | Distribution transformer overload risk assessment method | |
Cortes-Robles et al. | Fast-training feedforward neural network for multi-scale power quality monitoring in power systems with distributed generation sources | |
CN114580829A (en) | Power utilization safety sensing method, equipment and medium based on random forest algorithm | |
CN110045237A (en) | Transformer state parametric data prediction technique and system based on drosophila algorithm optimization | |
CN109934469A (en) | Based on the heterologous power failure susceptibility method for early warning and device for intersecting regression analysis | |
CN111680712B (en) | Method, device and system for predicting oil temperature of transformer based on similar time in day | |
CN110084502A (en) | A kind of power quality controlling equipment running status appraisal procedure and device | |
CN115829145A (en) | Photovoltaic power generation capacity prediction system and method | |
de Aquino et al. | A fuzzy system for detection of incipient fault in power transformers based on gas-in-oil analysis | |
Niazazari et al. | Event cause analysis in distribution networks using synchro waveform measurements | |
MansourLakouraj et al. | Event classification in active distribution grids using physics-informed graph neural network and PMU measurements | |
De et al. | Classification of power system voltage stability conditions using Kohonen's self‐organising feature map and learning vector quantisation | |
Malik et al. | Probabilistic neural network based incipient fault identification using DGA dataset | |
CN116956702A (en) | Electricity safety early warning method, medium and system | |
CN110378610A (en) | Distribution weak link identification method based on user's different degree and equipment running status | |
Qian et al. | Research on power transformer fault prediction model based on lstm neural network | |
Yong et al. | Research on condition evaluation algorithm of oil-immersed transformer based on Naive Bayes | |
CN114626433A (en) | Fault prediction and classification method, device and system for intelligent electric energy meter |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191112 |
|
RJ01 | Rejection of invention patent application after publication |