CN112329335A - Long-term prediction method for content of dissolved gas in transformer oil - Google Patents
Long-term prediction method for content of dissolved gas in transformer oil Download PDFInfo
- Publication number
- CN112329335A CN112329335A CN202011139435.6A CN202011139435A CN112329335A CN 112329335 A CN112329335 A CN 112329335A CN 202011139435 A CN202011139435 A CN 202011139435A CN 112329335 A CN112329335 A CN 112329335A
- Authority
- CN
- China
- Prior art keywords
- long
- prediction
- content
- data
- transformer oil
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 63
- 230000007774 longterm Effects 0.000 title claims abstract description 22
- 238000013528 artificial neural network Methods 0.000 claims abstract description 60
- 238000012549 training Methods 0.000 claims abstract description 39
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 30
- 230000006403 short-term memory Effects 0.000 claims abstract description 23
- 230000007787 long-term memory Effects 0.000 claims abstract description 22
- 230000002068 genetic effect Effects 0.000 claims abstract description 21
- 230000002779 inactivation Effects 0.000 claims abstract description 10
- 230000015654 memory Effects 0.000 claims description 50
- 230000008569 process Effects 0.000 claims description 16
- 238000004364 calculation method Methods 0.000 claims description 14
- 238000012360 testing method Methods 0.000 claims description 13
- 238000013178 mathematical model Methods 0.000 claims description 12
- 230000006870 function Effects 0.000 claims description 11
- 239000011159 matrix material Substances 0.000 claims description 10
- 230000001537 neural effect Effects 0.000 claims description 7
- 238000012545 processing Methods 0.000 claims description 7
- 230000000306 recurrent effect Effects 0.000 claims description 7
- 230000004913 activation Effects 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 5
- 238000012163 sequencing technique Methods 0.000 claims description 3
- 230000008859 change Effects 0.000 abstract description 4
- 239000007789 gas Substances 0.000 description 92
- VNWKTOKETHGBQD-UHFFFAOYSA-N methane Chemical compound C VNWKTOKETHGBQD-UHFFFAOYSA-N 0.000 description 10
- HSFWRNGVRCDJHI-UHFFFAOYSA-N Acetylene Chemical compound C#C HSFWRNGVRCDJHI-UHFFFAOYSA-N 0.000 description 6
- OTMSDBZUPAUEDD-UHFFFAOYSA-N Ethane Chemical compound CC OTMSDBZUPAUEDD-UHFFFAOYSA-N 0.000 description 6
- 238000011161 development Methods 0.000 description 5
- VGGSQFUCUMXWEO-UHFFFAOYSA-N Ethene Chemical compound C=C VGGSQFUCUMXWEO-UHFFFAOYSA-N 0.000 description 4
- 239000005977 Ethylene Substances 0.000 description 4
- UFHFLCQGNIYNRP-UHFFFAOYSA-N Hydrogen Chemical compound [H][H] UFHFLCQGNIYNRP-UHFFFAOYSA-N 0.000 description 4
- 125000002534 ethynyl group Chemical group [H]C#C* 0.000 description 4
- 239000001257 hydrogen Substances 0.000 description 4
- 229910052739 hydrogen Inorganic materials 0.000 description 4
- 238000013473 artificial intelligence Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 230000002411 adverse Effects 0.000 description 2
- 238000004587 chromatography analysis Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 230000003449 preventive effect Effects 0.000 description 2
- 238000012300 Sequence Analysis Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000012806 monitoring device Methods 0.000 description 1
- 230000035772 mutation Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000007637 random forest analysis Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/12—Computing arrangements based on biological models using genetic models
- G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/06—Electricity, gas or water supply
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2111/00—Details relating to CAD techniques
- G06F2111/10—Numerical modelling
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2113/00—Details relating to the application field
- G06F2113/08—Fluids
-
- 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
Abstract
The invention relates to a long-term prediction method of the content of dissolved gas in transformer oil, which optimizes data time window steps, network hidden layers, training times and random inactivation parameters by using a genetic algorithm and solves the problem of low precision of prediction results of a prediction model; compared with the traditional prediction model, the prediction model combining the long-term and short-term memory artificial neural network and the genetic algorithm has higher long-term prediction accuracy, and can better track the change trend of dissolved gas in oil.
Description
Technical Field
The invention belongs to the technical field of power system monitoring, and particularly relates to a long-term prediction method for content of dissolved gas in transformer oil.
Background
The power transformer is one of the most critical devices of the power system, and the stable, safe and fault-free operation of the power transformer is a key factor for ensuring the reliable power supply of the whole power system.
The dissolved gas in the transformer oil mainly contains H2、CH4、C2H6、C2H2、C2H4、CO、CO2The development process of the gas content of the seven gases has a fluctuation characteristic and is influenced by factors such as oil temperature, gas partial pressure of oil, fault property and development speed of the oil, and the dissolved gas content sequence in the transformer oil is also influenced by reasons such as load, operation time and the like, so that the gas content data does not change according to an exponential growth rule strictly, and the prediction effect can be influenced to a greater extent by the nonlinear and non-stationary characteristics of the gas content sequence in the transformer oil to a certain extent. The concentration data of the dissolved gas in the transformer oil with defects and faults also has the characteristics of small gas data sample and less information, and the characteristic determines that the concentration of the dissolved gas in the transformer oil can not be accurately predicted by the existing prediction technology. Therefore, it is necessary to improve the prediction accuracy by an appropriate method.
Historical gas monitoring sequences can be formed by utilizing the online oil chromatography monitoring device, and then related gases can be predicted. The current state of the transformer or some latent faults and the development degree thereof can be known through the content predicted by the dissolved gas in the transformer oil, and the method can be used as an important basis for evaluating the state of the transformer and has important reference significance for customizing a special preventive scheme under the adverse development trend of the transformer.
The existing oil chromatography prediction model generally adopts a mathematical statistical method to perform time sequence analysis on the dissolved gas data of the transformer, for example, a grey model is used for performing time sequence simulation on the gas data on the basis of an exponential curve; and modeling the gas content change trend by using a differential autoregressive moving average method. But the accuracy rate mainly depends on the distribution characteristics of the experimental data set. In the aspect of prediction, the complexity of the established model is low due to the fact that the fitting function of the difference autoregressive moving average method is too single, and the prediction result is difficult to reflect the actual situation.
The combined prediction model adopts different methods to predict the same problem, and has two basic forms: equal-weight combinations and unequal-weight combinations. The method generally adopted is unequal weight combination. The idea of combined prediction is mainly to fully utilize key information in each single prediction method and integrate prediction advantages of each single prediction method, so that the prediction accuracy is improved, and the problems of low prediction accuracy and the like possibly caused by single prediction method are solved. The difficulty in building a combined prediction model is that it is difficult to determine the weighting coefficients between different methods. For the existing combined prediction model, a prediction model with minimized construction error is generally adopted to calculate the optimal weighting coefficient. Although a good prediction effect is obtained, only information of a single prediction method is considered, and whether repeated information occurs between different prediction methods or not is not considered, so that the phenomenon of information redundancy is caused.
The artificial intelligence prediction utilizes devices such as sensors and the like to acquire data, processes and analyzes the data through a computer technology, and constructs a model for prediction. Common artificial intelligence methods include: random forest, support vector machine and recurrent neural network. Because the traditional artificial intelligence algorithm has defects in the problem of processing time sequences, a predicted value has a large error with an actual value, and the recurrent neural network overcomes the problem, but has the problems of short-term memory, incapability of processing long sequences and overlong training time. The long-short term memory artificial neural network is a special recurrent neural network model, and a memory module is added in the recurrent neural network, so that the long-short term memory artificial neural network is widely applied to the time series prediction problem. Relevant parameters of data time window step, network hiding layer, training times and random inactivation parameter (Dropout) are set only by means of manual experience, uncertainty exists, and the prediction effect of the model is reduced.
Disclosure of Invention
The invention aims to provide a long-term prediction method for the content of dissolved gas in transformer oil, which optimizes data time window steps, network hidden layers, training times and random inactivation parameters and solves the problem of low accuracy of prediction results of a prediction model.
The technical scheme adopted by the invention is that the long-term prediction method of the content of the dissolved gas in the transformer oil is implemented according to the following steps:
step 2, establishing a mathematical model by a long-term and short-term memory artificial neural network method;
step 3, searching the optimal parameters in the mathematical model by using a genetic algorithm to obtain an optimized long-term and short-term memory artificial neural network model;
and 5, inputting the content of the dissolved gas in the transformer oil as input data into the optimal long-term and short-term memory artificial neural network model to obtain a predicted value of the content of the dissolved gas in the transformer oil at the next stage.
The invention is also characterized in that:
in the step 1, the ratio of the test set to the training set is 1: 4.
The specific process of the step 2 is as follows: data of the content of dissolved gas in the transformer oil is participated in a long-term and short-term memory artificial neural network link and transmitted by adopting a cell state; the method specifically comprises the following steps:
at the time t, the long-short term memory artificial neural network method establishes a mathematical model, namely a cell state calculation formula as follows:
it＝σ(Wxixt+Whiht-1+bi)
ft＝σ(Wxfxt+Wfhht-1+bf)
ot＝σ(Woxxt+Wohht-1+bo)
c′t＝tanh(Wc·ht-1+Wf·xt+bc)
ct＝ft⊙ct-1+it⊙c′t
ht＝ot⊙tanh(ct)
wherein itRepresenting the state of the memory unit reserved for the content of the dissolved gas in the current transformer oil; otRepresenting the output of the gas content state in the control memory unit to the current state; f. oftRepresenting the gas content state in the memory unit from the last moment to the next moment; t represents a time; i denotes an input gate of the gas prediction model, f denotes a forgetting gate of the gas prediction model, o denotes an output gate of the gas prediction model, and Whi、WfhWho represents a weight matrix of the gas prediction model, and h represents a hidden layer; wxi、Wxf、WoxAll represent weight matrix, x represents input corresponding gas content; bi、bf、boAll represent bias terms, b is a coefficient; sigma represents a sigmoid activation function; c'tA cell state input representing time t; tan h is a hyperbolic tangent activation function; wcState weight matrix representing input layer, bcThe state bias term representing the input layer, C represents the memory cell state.
And 3, the optimal parameters in the mathematical model in the step 3 refer to the step length of a dissolved gas content data set in the transformer oil, a network hidden layer, training times and random inactivation.
The specific process of the step 4 is as follows:
inputting the training set into an optimized long-short term memory artificial neural network model for training;
inputting test set data of the content of dissolved gas in the transformer oil into a hidden layer of a trained optimized long-short term memory artificial neural network model, wherein a prediction model unit is influenced by a previous-stage training model, and outputting prediction data as follows:
p＝{p1，p2，p3，……，ps}
Pp＝LSTM-GAcal{Xp，Cp-1，Hp-1}
in the formula, Cp-1The state of the artificial neural network-genetic algorithm unit is memorized for the previous long and short term; hp-1Outputting the gas content of the previous long-short term memory artificial neural network-genetic algorithm unit; c is a memory unit for gas data; h is a hidden layer; p is the output gas prediction data; x is a training dataset for gas; s is the number of parts of a dissolved gas content data set in the transformer oil; LSTM-GAcal is the calculation process of the recurrent neural network;
performing error calculation on the output prediction data, and when the error is within a preset range, taking the trained optimized long-short term memory artificial neural network model as an optimal long-short term memory artificial neural network model; otherwise, modifying parameters in the genetic algorithm and returning to the step 3.
The specific process of error calculation on the output prediction data comprises the following steps:
selecting actual data from a training set of the content of dissolved gas in transformer oil to form a number set Yf：
Yf＝{dm-S+1,dm-S+2,……，dm}
In the formula: f represents the original sequence, dmThe normalized value of the content of the dissolved gas in the transformer oil at the time m is obtained; s is a step length for dividing a dissolved gas content data set in the transformer oil;
using long-and-short term memory artifactsModel pair Y combining neural network and genetic algorithmfAnd (4) predicting, and outputting the following results:
Pf＝LSTM-GA(Yf)＝{Pm-S+2,Pm-S+3,……，Pm+1}
in the formula: pm+1Representing the predicted value at time m +1, and calculating YfThe gas data of (1) is eliminated, and P is addedm+1The new data formed by combination is:
Yf+1＝{dm-S+2,dm-S+3,……,Pm+1}
and a radical of Yf+1Inputting the data into a prediction model combining a long-term and short-term memory artificial neural network and a genetic algorithm to obtain Pm+2And iterating to obtain a pre-sequencing column as follows:
Pte＝{Pm+1,Pm+2,……，Pn}
in the formula: te represents the latest sequence, and n is the sequence number;
inputting the prediction data in the prediction sequence and actual data corresponding to the same time into a root mean square error formula to obtain prediction precision;
the root mean square error formula is:
in the formula, eRMSEIndicating the accuracy of the prediction, yiThe actual data is represented by a representation of,
representing the prediction data.
The long-term prediction method for the content of the dissolved gas in the transformer oil has the beneficial effects that:
the method for predicting the dissolved gas in the oil by combining the long-term and short-term neural network with the genetic algorithm is provided, the genetic algorithm is used for optimizing a data time window step, a network hidden layer, training times and random inactivation parameters, and the problem of low accuracy of a prediction result of a prediction model is solved;
compared with the traditional prediction model, the prediction model combining the long-term and short-term memory artificial neural network and the genetic algorithm has higher long-term prediction accuracy, and can better track the change trend of dissolved gas in oil.
Drawings
FIG. 1 is a flow chart of a long-term prediction method for the content of dissolved gases in transformer oil according to the present invention;
FIG. 2 is a graph of methane prediction results for an embodiment of the present invention;
FIG. 3 is a graph illustrating ethane prediction results for an embodiment of the present invention;
FIG. 4 is a graphical illustration of acetylene prediction results for an embodiment of the invention;
FIG. 5 is a graph illustrating ethylene predictions for an embodiment of the present invention;
FIG. 6 is a diagram illustrating hydrogen prediction results according to an embodiment of the present invention.
Detailed Description
The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.
The invention discloses a long-term prediction method for the content of dissolved gas in transformer oil, which is implemented according to the following steps as shown in figure 1:
the ratio of the test set to the training set is 1: 4.
Because the problems of high prediction difficulty and low precision of the content of the dissolved gas in the transformer oil exist, in order to improve the prediction precision of the content of the dissolved gas in the transformer oil and the calculation speed of obtaining a prediction result, the minimum-maximum normalization processing needs to be carried out on the sample content of each gas, the sample content is mapped between [0 and 1], and the conversion function is as follows:
in the formula: x is the number of*The processed gas data; x is the gas data to be processed; x is the number ofminIs the minimum in the gas data; x is the number ofmaxIs the maximum in the gas data; min represents the minimum value; max represents the maximum value.
Step 2, establishing a mathematical model by a long-term and short-term memory artificial neural network method;
the specific process of the step 2 is as follows: data of the content of dissolved gas in the transformer oil is participated in a long-term and short-term memory artificial neural network link and transmitted by adopting a cell state; the method specifically comprises the following steps:
at the time t, the long-short term memory artificial neural network method establishes a mathematical model, namely a cell state calculation formula as follows:
it＝σ(Wxixt+Whiht-1+bi)
ft＝σ(Wxfxt+Wfhht-1+bf)
ot＝σ(Woxxt+Wohht-1+bo)
c′t＝tanh(Wc·ht-1+Wf·xt+bc)
ct＝ft⊙ct-1+it⊙c′t
ht＝ot⊙tanh(ct)
wherein itRepresenting the state of the memory unit reserved for the content of the dissolved gas in the current transformer oil; otRepresenting the output of the gas content state in the control memory unit to the current state; f. oftRepresenting the gas content state in the memory unit from the last moment to the next moment; t represents a time; i denotes an input gate of the gas prediction model, f denotes a forgetting gate of the gas prediction model, o denotes an output gate of the gas prediction model, and Whi、WfhWho represents a weight matrix of the gas prediction model, and h represents a hidden layer; wxi、Wxf、WoxAll represent weight matrix, x represents input corresponding gas content; bi、bf、boAll represent bias terms, b is a coefficient; sigma represents a sigmoid activation function; c. Ct' represents the cell state input at time t; tan h is a hyperbolic tangent activation function; wcState weight matrix representing input layer, bcThe state bias term representing the input layer, C represents the memory cell state.
When the long and short term memory network trains the content data of the dissolved gas in the transformer oil, calculating the output value of the gas of the long and short term memory network unit according to forward propagation, calculating the error value between the gas output value and the true value of the long and short term memory network unit and performing backward propagation; calculating a weight gradient according to an error value of the content of the dissolved gas in the transformer oil; and gradient descending is carried out, and the weight value is returned and updated in real time.
Step 3, searching the optimal parameters in the mathematical model by using a genetic algorithm to obtain an optimized long-term and short-term memory artificial neural network model;
the optimal parameters in the mathematical model refer to step length, network hidden layer, training times and random inactivation of a dissolved gas content data set in the transformer oil.
The specific process of the step 3 is as follows:
step 3.1, initializing and decoding the population;
step 3.2, the mean square error of the long-term and short-term memory artificial neural network is used as a fitness function;
3.3, carrying out selection cross variation operation on the solved individuals;
step 3.4, if the target value of the fitness function reaches the optimal value, carrying out the next step; otherwise, returning to the step 3.3;
step 3.5, acquiring a fitness target value and an optimal parameter;
step 3.6, calculating the predicted mean square error based on the optimal parameters;
step 3.7, judging termination conditions, stopping calculation if the number of times of population iteration is met, and outputting the step length of the dissolved gas content data set in the long-term and short-term memory artificial neural network global optimum parameter combination transformer oil, a network hidden layer, training times and random inactivation (Dropout) parameters; otherwise, returning to the step 3.6.
the specific process of the step 4 is as follows:
inputting the training set into an optimized long-short term memory artificial neural network model for training;
inputting test set data of the content of dissolved gas in the transformer oil into a hidden layer of a trained optimized long-short term memory artificial neural network model, wherein a prediction model unit is influenced by a previous-stage training model, and outputting prediction data as follows:
p＝{p1,p2,p3,……，ps}
Pp＝LSTM-GAcal{Xp,Cp-1,Hp-1}
in the formula, Cp-1The state of the artificial neural network-genetic algorithm unit is memorized for the previous long and short term; hp-1Outputting the gas content of the previous long-short term memory artificial neural network-genetic algorithm unit; c is a memory unit for gas data; h is a hidden layer; p is the output gas prediction data; x is a training dataset for gas; s is the number of parts of a dissolved gas content data set in the transformer oil; LSTM-GAcal is the calculation process of the recurrent neural network;
the parameters selected by the genetic algorithm are adopted in the long-short term memory artificial neural network-genetic algorithm combined model, and the optimization target is the minimization of a loss function.
Performing error calculation on the output prediction data, and when the error is within a preset range, taking the trained optimized long-short term memory artificial neural network model as an optimal long-short term memory artificial neural network model; otherwise, modifying parameters in the genetic algorithm, returning to the step 3, and modifying the parameters in the genetic algorithm, including the population size; termination of evolution algebra, cross probability and mutation probability.
The specific process of error calculation on the output prediction data comprises the following steps:
selecting actual data from a training set of the content of dissolved gas in transformer oil to form a number set Yf：
Yf＝{dm-S+1,dm-S+2,……，dm}
In the formula: f represents the original sequence, dmThe normalized value of the content of the dissolved gas in the transformer oil at the time m is obtained; s is a step length for dividing a dissolved gas content data set in the transformer oil;
model pair Y combining long-short term memory artificial neural network and genetic algorithmfAnd (4) predicting, and outputting the following results:
Pf＝LSTM-GA(Yf)＝{Pm-S+2,Pm-S+3,……，Pm+1}
in the formula: pm+1Representing the predicted value at time m +1, and calculating YfThe gas data of (1) is eliminated, and P is addedm+1The new data formed by combination is:
Yf+1＝{dm-S+2,dm-S+3,……,Pm+1}
and a radical of Yf+1Inputting the data into a prediction model combining a long-term and short-term memory artificial neural network and a genetic algorithm to obtain Pm+2And iterating to obtain a pre-sequencing column as follows:
Pte＝{Pm+1,Pm+2,……，Pn}
in the formula: te represents the latest sequence, and n is the sequence number;
inputting the prediction data in the prediction sequence and actual data corresponding to the same time into a root mean square error formula to obtain prediction precision;
the root mean square error formula is:
in the formula, eRMSEIndicating the accuracy of the prediction, yiThe actual data is represented by a representation of,
representing the prediction data.
And 5, inputting the content of the dissolved gas in the transformer oil as input data into the optimal long-term and short-term memory artificial neural network model to obtain a predicted value of the content of the dissolved gas in the transformer oil at the next stage.
The invention relates to a long-term prediction method for the content of dissolved gas in transformer oil, wherein the method comprises the following steps: determining the content of dissolved gas in 5 kinds of transformer oil in a certain time period; sample data of the content of dissolved gas in the transformer oil is subjected to minimum-maximum normalization processing; differentiating data into a test set and a training set; optimizing four parameters of a data time window step, a network hidden layer in a long-short term memory artificial neural network model, training times and random inactivation (Dropout) according to the processes of encoding, population initialization, individual fitness evaluation, selection, crossing and variation in the population of a genetic algorithm, and judging whether the four parameters are optimal solutions; transmitting the four parameters of the determined data time window step, the network hidden layer, the training times and the random inactivation (Dropout) to the long-short term memory artificial neural network; the method comprises the steps of taking the content of dissolved gas in the transformer oil as input data, taking a predicted value of the content of the dissolved gas in the transformer oil at the next stage as an output matrix, iteratively and adaptively adjusting model weight through a long-short term memory cyclic neural network, fitting to form a prediction model combining a long-short term memory artificial neural network and a genetic algorithm, and outputting the predicted value of the content of the long-term gas according to the model.
Examples
The method and the BP neural network are adopted for predicting methane, ethane, acetylene, ethylene and hydrogen in the transformer respectively, the prediction result of the methane is shown in figure 2, the prediction result of the ethane is shown in figure 3, the prediction result of the acetylene is shown in figure 4, the prediction result of the ethylene is shown in figure 5, and the prediction result of the hydrogen is shown in figure 6; in fig. 2-6, the curves predicted by the method of the present invention and the curves predicted by the BP neural network are compared with the actual data curves, respectively, and it can be seen that the curves predicted by the method of the present invention are closer to the actual data curves.
The errors of the predictions of methane, ethane, acetylene, ethylene and hydrogen in the transformer by using the method and the BP neural network are shown in Table 1.
TABLE 1
| GAS | CH4 | C2H6 | C2H2 | C2H4 | H2 |
| LSMT-GA | 0.0728 | 0.0125 | 0.0207 | 0.0467 | 0.0325 |
| BP | 0.1042 | 0.0482 | 0.0530 | 0.0587 | 0.0551 |
As can be seen from the comparison in Table 1, the method of the invention has smaller error of the predicted result, and the predicted result is more accurate compared with the BP neural network algorithm.
Through the mode, the long-term prediction method for the content of the dissolved gas in the transformer oil can be used as an important basis for evaluating the state of the transformer according to the long-term prediction value of the content of the dissolved gas in the transformer oil output by the model, has extremely important reference significance for a customizable exclusive preventive scheme under the adverse development trend of the transformer, and prolongs the service life of the transformer. And meanwhile, the negative influences on the stability, reliability and economy of the power system caused by transformer faults are prevented.
Claims (6)
1. A long-term prediction method for the content of dissolved gas in transformer oil is characterized by comprising the following steps:
step 1, measuring the content of dissolved gas in transformer oil as sample data, performing minimum-maximum normalization processing, and differentiating the normalized data into a test set and a training set;
step 2, establishing a mathematical model by a long-term and short-term memory artificial neural network method;
step 3, searching the optimal parameters in the mathematical model by using a genetic algorithm to obtain an optimized long-term and short-term memory artificial neural network model;
step 4, performing network training on the optimized long-short term memory artificial neural network model through a training set, and detecting through a test set to obtain an optimal long-short term memory artificial neural network model;
and 5, inputting the content of the dissolved gas in the transformer oil as input data into the optimal long-term and short-term memory artificial neural network model to obtain a predicted value of the content of the dissolved gas in the transformer oil at the next stage.
2. The method for long-term prediction of the content of the dissolved gas in the transformer oil according to claim 1, wherein the ratio of the test set to the training set in step 1 is 1: 4.
3. The method for long-term prediction of the content of the dissolved gas in the transformer oil according to claim 1, wherein the specific process in the step 2 is as follows: data of the content of dissolved gas in the transformer oil is participated in a long-term and short-term memory artificial neural network link and transmitted by adopting a cell state; the method specifically comprises the following steps:
at the time t, the long-short term memory artificial neural network method establishes a mathematical model, namely a cell state calculation formula as follows:
it＝σ(Wxixt+Whiht-1+bi)
ft＝σ(Wxfxt+Wfhht-1+bf)
ot＝σ(Woxxt+Wohht-1+bo)
c′t＝tanh(Wc·ht-1+Wf·xt+bc)
ct＝ft⊙ct-1+it⊙c′t
ht＝ot⊙tanh(ct)
wherein itRepresenting the state of the memory unit reserved for the content of the dissolved gas in the current transformer oil; otRepresenting the output of the gas content state in the control memory unit to the current state; f. oftRepresenting the gas content state in the memory unit from the last moment to the next moment; t represents a time; i denotes an input gate of the gas prediction model, f denotes a forgetting gate of the gas prediction model, o denotes an output gate of the gas prediction model, and Whi、WfhWho represents a weight matrix of the gas prediction model, and h represents a hidden layer; wxi、Wxf、WoxAll represent weight matrix, x represents input corresponding gas content; bi、bf、boAll represent bias terms, b is a coefficient; sigma represents a sigmoid activation function; c'tRepresenting time tInputting a unit state; tan h is a hyperbolic tangent activation function; wcState weight matrix representing input layer, bcThe state bias term representing the input layer, C represents the memory cell state.
4. The method for long-term prediction of the content of the dissolved gas in the transformer oil according to claim 1, wherein the optimal parameters in the mathematical model in the step 3 refer to step length of a dissolved gas content data set in the transformer oil, a network hidden layer, training times and random inactivation parameters.
5. The method for long-term prediction of the content of the dissolved gas in the transformer oil according to claim 1, wherein the specific process in the step 4 is as follows:
inputting the training set into an optimized long-short term memory artificial neural network model for training;
inputting test set data of the content of dissolved gas in the transformer oil into a hidden layer of a trained optimized long-short term memory artificial neural network model, wherein a prediction model unit is influenced by a previous-stage training model, and outputting prediction data as follows:
p＝{p1,p2,p3,……，ps}
Pp＝LSTM-GAcal{Xp,Cp-1,Hp-1}
in the formula, Cp-1The state of the artificial neural network-genetic algorithm unit is memorized for the previous long and short term; hp-1Outputting the gas content of the previous long-short term memory artificial neural network-genetic algorithm unit; c is a memory unit for gas data; h is a hidden layer; p is the output gas prediction data; x is a training dataset for gas; s is the number of parts of a dissolved gas content data set in the transformer oil; LSTM-GAcal is the calculation process of the recurrent neural network;
performing error calculation on the output prediction data, and when the error is within a preset range, taking the trained optimized long-short term memory artificial neural network model as an optimal long-short term memory artificial neural network model; otherwise, modifying parameters in the genetic algorithm and returning to the step 3.
6. The method for long-term prediction of the content of the dissolved gas in the transformer oil according to claim 5, wherein the specific process of error calculation of the output prediction data is as follows:
selecting actual data from a training set of the content of dissolved gas in transformer oil to form a number set Yf：
Yf＝{dm-S+1,dm-S+2,……，dm}
In the formula: f represents the original sequence, dmThe normalized value of the content of the dissolved gas in the transformer oil at the time m is obtained; s is a step length for dividing a dissolved gas content data set in the transformer oil;
model pair Y combining long-short term memory artificial neural network and genetic algorithmfAnd (4) predicting, and outputting the following results:
Pf＝LSTM-GA(Yf)＝{Pm-S+2,Pm-S+3,……，Pm+1}
in the formula: pm+1Representing the predicted value at time m +1, and calculating YfThe gas data of (1) is eliminated, and P is addedm+1The new data formed by combination is:
Yf+1＝{dm-S+2,dm-S+3,……,Pm+1}
and a radical of Yf+1Inputting the data into a prediction model combining a long-term and short-term memory artificial neural network and a genetic algorithm to obtain Pm+2And iterating to obtain a pre-sequencing column as follows:
Pte＝{Pm+1,Pm+2,……，Pn}
in the formula: te represents the latest sequence, and n is the sequence number;
inputting the prediction data in the prediction sequence and actual data corresponding to the same time into a root mean square error formula to obtain prediction precision;
the root mean square error formula is:
in the formula, eRMSEIndicating the accuracy of the prediction, yiThe actual data is represented by a representation of,
representing the prediction data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011139435.6A CN112329335B (en) | 2020-10-22 | 2020-10-22 | Long-term prediction method for content of dissolved gas in transformer oil |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011139435.6A CN112329335B (en) | 2020-10-22 | 2020-10-22 | Long-term prediction method for content of dissolved gas in transformer oil |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112329335A true CN112329335A (en) | 2021-02-05 |
| CN112329335B CN112329335B (en) | 2024-03-01 |
Family
ID=74310972
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011139435.6A Active CN112329335B (en) | 2020-10-22 | 2020-10-22 | Long-term prediction method for content of dissolved gas in transformer oil |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112329335B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114935627A (en) * | 2022-05-07 | 2022-08-23 | 金川集团信息与自动化工程有限公司 | Method for predicting purity of hydrogen chloride at outlet of synthesis furnace |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109243172A (en) * | 2018-07-25 | 2019-01-18 | 华南理工大学 | Traffic flow forecasting method based on genetic algorithm optimization LSTM neural network |
| CN110441500A (en) * | 2019-08-15 | 2019-11-12 | 昆明理工大学 | A kind of predicting model for dissolved gas in transformer oil method based on shot and long term memory network |
| CN110501585A (en) * | 2019-07-12 | 2019-11-26 | 武汉大学 | A kind of Diagnosis Method of Transformer Faults based on Bi-LSTM and dissolved gas analysis |
| 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 |
-
2020
- 2020-10-22 CN CN202011139435.6A patent/CN112329335B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109243172A (en) * | 2018-07-25 | 2019-01-18 | 华南理工大学 | Traffic flow forecasting method based on genetic algorithm optimization LSTM neural network |
| CN110501585A (en) * | 2019-07-12 | 2019-11-26 | 武汉大学 | A kind of Diagnosis Method of Transformer Faults based on Bi-LSTM and dissolved gas analysis |
| CN110441500A (en) * | 2019-08-15 | 2019-11-12 | 昆明理工大学 | A kind of predicting model for dissolved gas in transformer oil method based on shot and long term memory network |
| 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 |
Non-Patent Citations (1)
| Title |
|---|
| 王科等: "基于LSTM 网络的变压器油中溶解气体浓度预测", 电子测量技术, vol. 43, no. 4, pages 1 - 2 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114935627A (en) * | 2022-05-07 | 2022-08-23 | 金川集团信息与自动化工程有限公司 | Method for predicting purity of hydrogen chloride at outlet of synthesis furnace |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112329335B (en) | 2024-03-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108900346B (en) | Wireless network flow prediction method based on LSTM network | |
| Benardos et al. | Optimizing feedforward artificial neural network architecture | |
| CN111860982A (en) | Wind power plant short-term wind power prediction method based on VMD-FCM-GRU | |
| CN111813084B (en) | Mechanical equipment fault diagnosis method based on deep learning | |
| CN109902801A (en) | A kind of flood DATA PROCESSING IN ENSEMBLE PREDICTION SYSTEM method based on variation reasoning Bayesian neural network | |
| Qu et al. | Genetic optimization method of pantograph and catenary comprehensive monitor status prediction model based on adadelta deep neural network | |
| CN111144552B (en) | Multi-index grain quality prediction method and device | |
| CN110705743A (en) | New energy consumption electric quantity prediction method based on long-term and short-term memory neural network | |
| CN112488235A (en) | Elevator time sequence data abnormity diagnosis method based on deep learning | |
| CN112364560B (en) | Intelligent prediction method for working hours of mine rock drilling equipment | |
| CN110824915A (en) | GA-DBN network-based intelligent monitoring method and system for wastewater treatment | |
| CN112085254A (en) | Prediction method and model based on multi-fractal cooperative measurement gating cycle unit | |
| CN114707712A (en) | Method for predicting requirement of generator set spare parts | |
| US20230237316A1 (en) | Forecast method and system of wind power probability density | |
| Zeng et al. | Short-term load forecasting of smart grid systems by combination of general regression neural network and least squares-support vector machine algorithm optimized by harmony search algorithm method | |
| CN113553768B (en) | Method and device for rapidly calculating reliability of power grid driven by model data in hybrid mode | |
| CN112329335B (en) | Long-term prediction method for content of dissolved gas in transformer oil | |
| CN109408896B (en) | Multi-element intelligent real-time monitoring method for anaerobic sewage treatment gas production | |
| CN110633868A (en) | Method for predicting properties of oil and gas in exploration well oil testing layer by optimizing neural network through genetic algorithm | |
| CN114021432A (en) | Stress corrosion cracking crack propagation rate prediction method and system | |
| CN116303786B (en) | Block chain financial big data management system based on multidimensional data fusion algorithm | |
| CN112862063A (en) | Complex pipe network leakage positioning method based on deep belief network | |
| CN117076887A (en) | Pump station unit running state prediction and health assessment method and system | |
| CN112232570A (en) | Forward active total electric quantity prediction method and device and readable storage medium | |
| CN115759343A (en) | E-LSTM-based user electric quantity prediction method and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |