CN115238754A - Power transformer short-term operation temperature prediction method based on multivariate perception - Google Patents
Power transformer short-term operation temperature prediction method based on multivariate perception Download PDFInfo
- Publication number
- CN115238754A CN115238754A CN202211150771.XA CN202211150771A CN115238754A CN 115238754 A CN115238754 A CN 115238754A CN 202211150771 A CN202211150771 A CN 202211150771A CN 115238754 A CN115238754 A CN 115238754A
- Authority
- CN
- China
- Prior art keywords
- feature
- data
- temperature
- power transformer
- load current
- 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 35
- 230000008447 perception Effects 0.000 title claims abstract description 34
- 238000012417 linear regression Methods 0.000 claims abstract description 23
- 230000004927 fusion Effects 0.000 claims abstract description 21
- 238000004804 winding Methods 0.000 claims abstract description 13
- 238000012216 screening Methods 0.000 claims abstract description 11
- 230000001052 transient effect Effects 0.000 claims abstract description 8
- 230000009466 transformation Effects 0.000 claims abstract description 7
- 238000007781 pre-processing Methods 0.000 claims abstract description 4
- 230000001364 causal effect Effects 0.000 claims description 24
- 238000003066 decision tree Methods 0.000 claims description 13
- 239000013598 vector Substances 0.000 claims description 13
- 238000012549 training Methods 0.000 claims description 11
- 238000000354 decomposition reaction Methods 0.000 claims description 9
- 238000000605 extraction Methods 0.000 claims description 8
- 230000006870 function Effects 0.000 claims description 7
- 230000008859 change Effects 0.000 claims description 6
- 238000013139 quantization Methods 0.000 claims description 6
- 238000005192 partition Methods 0.000 claims description 5
- 230000008569 process Effects 0.000 claims description 5
- 239000000126 substance Substances 0.000 claims description 5
- 230000004913 activation Effects 0.000 claims description 4
- 238000010606 normalization Methods 0.000 claims description 4
- 238000004458 analytical method Methods 0.000 claims description 3
- 238000004587 chromatography analysis Methods 0.000 claims description 3
- 238000001816 cooling Methods 0.000 claims description 3
- 238000012937 correction Methods 0.000 claims description 3
- 238000011478 gradient descent method Methods 0.000 claims description 3
- 230000007246 mechanism Effects 0.000 claims description 3
- 230000011218 segmentation Effects 0.000 claims description 3
- 230000003044 adaptive effect Effects 0.000 claims description 2
- 230000009849 deactivation Effects 0.000 claims description 2
- 230000002159 abnormal effect Effects 0.000 abstract description 3
- 238000012544 monitoring process Methods 0.000 abstract description 2
- 230000010339 dilation Effects 0.000 description 6
- 238000010586 diagram Methods 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 238000009529 body temperature measurement Methods 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 239000013307 optical fiber Substances 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 230000002779 inactivation Effects 0.000 description 1
- 238000009413 insulation Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000013021 overheating Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
Images
Classifications
-
- 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
-
- 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—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Economics (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Biophysics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Tourism & Hospitality (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Evolutionary Computation (AREA)
- Data Mining & Analysis (AREA)
- Game Theory and Decision Science (AREA)
- Development Economics (AREA)
- Water Supply & Treatment (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Operations Research (AREA)
- Public Health (AREA)
- Quality & Reliability (AREA)
- Primary Health Care (AREA)
- Entrepreneurship & Innovation (AREA)
Abstract
The invention belongs to the technical field of power transformer monitoring, and relates to a power transformer short-term operation temperature prediction method based on multivariate perception, which comprises the steps of obtaining historical state perception data of a power transformer, preprocessing the state perception data by using wavelet transformation, and removing noise signals in the state perception data; performing feature screening based on the state perception data, and constructing an original feature data set for predicting the hotspot temperature transient curve; constructing a multiple linear regression prediction model for the future short-term load current, and determining a final predicted value of the load current by integrating the multiple linear regression prediction model and the predicted value of the power dispatching system; and constructing a time convolution network prediction model based on multi-feature scale fusion according to the original feature data set, and predicting the space oil temperature and the winding temperature. The invention can effectively utilize and analyze the multidimensional power transformer state data and effectively predict the operating temperature of the power transformer in an abnormal state.
Description
Technical Field
The invention belongs to the technical field of power transformer monitoring, and particularly relates to a power transformer short-term operation temperature prediction method based on multivariate perception.
Background
The power transformer is the core and hub device of the power system, and the operation state thereof directly affects the safety and stability of the power system. The operation temperature of the power transformer mainly comprises oil level temperature and winding temperature, and the change of the operation temperature of the power transformer can reflect the internal overheating and insulation damage faults of the transformer to a certain degree. Therefore, historical operating temperature change and state sensing data of the power transformer are collected, and short-term prediction of the operating temperature is carried out, so that the method has important significance on the operation safety of a power grid and the power transformer.
The invention patent of China with application number 202111415692.2 discloses a short-term prediction method and a prediction system for top layer oil temperature of a distribution transformer, which predicts the short-term oil temperature by acquiring historical data of the top layer oil temperature of the distribution transformer, processing the historical data into time series data and applying an ISSA algorithm. The invention patent of China with the application number of 201811464867.7 discloses a method for predicting the hot-spot temperature rise of a transformer by comparing optical fiber temperature measurement, and the method comprises the steps of obtaining data of the hot-spot temperature, the top-layer oil temperature, the environment temperature and the load current measured by an oil-immersed transformer provided with optical fiber temperature measurement equipment, establishing an explicit hot-spot temperature rise model and predicting the hot-spot temperature rise of the transformer.
In the prior art, the transformer temperature rise is mainly predicted based on the historical temperature of the power transformer, however, the operating temperature of the power transformer is influenced by multiple factors, and when the power transformer has internal faults, multi-dimensional abnormal state changes including temperature rise, vibration, noise and the like can occur.
Disclosure of Invention
In order to solve the technical problem, the invention provides a power transformer short-term operation temperature prediction method based on multivariate perception.
The technical scheme adopted by the invention is as follows: a power transformer short-term operation temperature prediction method based on multivariate perception comprises the following steps:
acquiring historical state sensing data of a power transformer, preprocessing the state sensing data by using wavelet transformation, and removing noise signals in the state sensing data;
secondly, performing feature screening based on the state perception data; firstly, manually selecting characteristic factors influencing the oil level and the hot spot temperature of the transformer; then, further screening out characteristic factors by using a decision tree, removing redundant data, and constructing an original characteristic data set for predicting the hotspot temperature transient curve;
constructing a multiple linear regression prediction model for the future short-term load current, and predicting the future short-term load current based on the historical load current, the holiday variables and the air temperature variables; meanwhile, calling a short-term load prediction curve of the power dispatching system, and determining a final prediction value of the load current by integrating a multiple linear regression prediction model and the prediction value of the power dispatching system;
and step four, according to the original characteristic data set, considering the space distribution and time delay characteristics of the operation temperature change of the transformer, constructing a time convolution network prediction model based on multi-characteristic scale fusion, and predicting the space oil temperature and the winding temperature.
Preferably, in the step one, the state sensing data of the power transformer includes oil level temperature, three-phase winding temperature, load current, bias current, three-phase current unbalance, ambient temperature, ambient wind power, radiator temperature difference deviation, cooling power, vibration and oil chromatography data.
Preferably, in the first step, according to wavelet multi-resolution analysis, a Symlet wavelet is selected as a wavelet base, 5 layers of wavelet decomposition are carried out on each dimension of historical data, threshold quantization is carried out on the obtained wavelet decomposition coefficients, an extreme value threshold estimation method is adopted for threshold selection, the wavelet decomposition coefficients are processed by a soft threshold denoising method, and finally wavelet inverse transformation reconstruction is carried out by combining the 5 th layer of low-frequency coefficients and the high-frequency coefficients of each layer subjected to threshold quantization processing to obtain denoised state sensing data.
Further preferably, in the step two, the CART decision tree is used for optimal characteristic factor selection: step S1: each sample in the state perception data sample set comprises data of each characteristic quantity manually selected in advance and a corresponding measured temperature value; calculating a kini coefficient of each feature according to sample data, selecting the feature with the minimum kini coefficient and a value corresponding to the feature as an optimal feature and an optimal segmentation point, and dividing a sample set into two subsets; step S2: repeating the step S1, recursively constructing a binary decision tree, and the step S3: and calculating the importance index of each feature according to the finally generated decision tree, eliminating the features with low correlation, completing feature factor screening, and constructing hot spot temperature transient curves of different structural members to predict an original feature data set.
Further preferably, in step three, the process of determining the final predicted value of the load current is as follows:
wherein，tAs a matter of time, the time is,i f (t)、i fa (t)、i fb (t) Respectively the final predicted value of the load current, the predicted value of the multiple linear regression model and the predicted value of the power dispatching system,pfor the accuracy penalty coefficients of the multiple linear regression model,qthe accuracy punishment coefficient of the power dispatching system comprises the following components:
wherein，i p (c)、i pa (c)、i pb (c) Respectively is a historical true value of the load current, a historical predicted value of the multiple linear regression model and a historical predicted value of the power dispatching system,cis a variable of the time, and is,nto predict the number of samples in time.
Preferably, in the third step, the historical load current, the variables of the holidays, the air temperature and the time are selected as characteristic quantitiesxThe multiple linear regression prediction model for the future short-term load current is as follows:
whereinβIs a vector of the coefficients of the regression,ξare random error terms, and are model parameters obtained by learning,x 1 in order to be able to history the load current,x 2 e {0,1} is a holiday variable, x 3 is a variable of the air temperature,x 4 is a variable of the time, and is,β 0 a regression coefficient that represents the amount of feature,β 1 a regression coefficient representing the historical load current,β 2 a regression coefficient representing a holiday variable,β 3 a regression coefficient representing a variable of the air temperature,β 4 expressing the regression coefficient of the time variable, and taking the square error as the model errorExpressed as:
wherein the content of the first and second substances,J(β)in order to be a model error,mthe number of samples input for training is the number of samples,iis a sample number, and is a sample number,h β (x i() )in order to predict the load current(s),x i() indicating a sample set with a sequence number ofiThe characteristic quantity of (a) is calculated,y i() indicating a sample set of sequence numbers ofiTraining and reducing the model error by using a gradient descent method, wherein the model parameter updating expression is as follows:
wherein, the first and the second end of the pipe are connected with each other,
the regression coefficient is updated for the jth,β j is the (j) th regression coefficient,αto update the step size.
Further preferably, the time convolution network prediction model based on multi-feature scale fusion comprises a feature data set reconstruction module, a feature extraction module, a feature fusion module and a multi-layer perceptron MLP prediction module which are sequentially arranged.
Preferably, the feature extraction module adopts a residual expansion causal convolution unit, the residual expansion causal convolution unit is formed by connecting two forward propagation modules through a residual, and each forward propagation module is formed by connecting an expansion causal convolution layer, a weight normalization layer, a correction linear unit layer and a random inactivation layer in sequence.
Further preferably, the time convolution network prediction model based on multi-feature scale fusion is constructed as follows:
step 4.1: constructing a characteristic data set reconstruction module: splicing the current time original characteristic data set data and the historical original characteristic data set data to form a reconstructed characteristic data set with different time scales, wherein the characteristic dimensions are described as follows:
whereind new Is the dimension of the reconstructed feature data set,d in is the dimension of the input original feature data set,d out is the dimension of the model output data,ωis a delay parameter;
step 4.2: a characteristic extraction module is constructed, a residual expansion causal convolution unit is adopted to extract space-time characteristic vectors according to the reconstruction characteristic data sets, and the reconstruction characteristic data sets corresponding to each space partition are respectively input into one residual expansion causal convolution unit to obtain corresponding space-time characteristic vectors;
step 4.3: a feature fusion module is constructed, and weighted fusion is carried out on the space-time feature vectors output by each residual error expansion causal convolution unit;
step 4.4: and inputting the fused data into a multi-layer perceptron MLP prediction module for prediction, and finally outputting predicted temperature distribution data.
Further preferably, in step 4.3, the weight of each residual error dilation causal convolution unit is automatically adjusted by using an adaptive mechanism, a one-dimensional residual error dilation causal convolution unit is used to obtain an initial weight, and a softmax function is selected as an activation function of the initial weight.
The invention has the technical effects that: further screening out characteristic factors by using a decision tree to construct an original characteristic data set for predicting the hot spot temperature transient curve; constructing a multiple linear regression prediction model for the future short-term load current, comprehensively comparing the multiple linear regression prediction model with the predicted value of the power dispatching system, and determining the final predicted value of the load current; and finally, a time convolution network prediction model based on multi-feature scale fusion is constructed for prediction, so that multi-dimensional power transformer state data can be effectively utilized and analyzed, and the operating temperature of the power transformer in an abnormal state can be effectively predicted.
Drawings
FIG. 1 is a flow chart of a power transformer short-term operation temperature prediction method based on multivariate perception.
FIG. 2 is a schematic diagram of a time convolution network prediction model based on multi-feature scale fusion.
Fig. 3 is a schematic diagram of a process of reconstructing an original feature data set.
FIG. 4 is a schematic diagram of a residual dilation causal convolution unit.
Detailed Description
The invention is explained in more detail below with reference to the figures and examples.
Referring to fig. 1, a power transformer short-term operation temperature prediction method based on multivariate perception comprises the following steps:
step one, acquiring historical state perception data of the power transformer, and preprocessing the state perception data by using wavelet transformation to remove noise signals in the state perception data.
Secondly, performing feature screening based on the state perception data; firstly, manually selecting characteristic factors which may influence the oil level and the hot spot temperature of the transformer, such as the oil level temperature, the temperature of a three-phase winding, load current, bias current and the like; and then, further screening out characteristic factors with higher influence degree by using a decision tree, removing redundant data, and constructing an original characteristic data set for predicting the hotspot temperature transient curve.
Constructing a multiple linear regression prediction model for the future short-term load current, and predicting the future short-term load current based on the historical load current, the holiday variables and the air temperature variables; meanwhile, calling a short-term load prediction curve of the power dispatching system, integrating the multiple linear regression prediction model and the predicted value of the power dispatching system, and determining the final predicted value of the load current:
wherein，tAs a matter of time, the time is,i f (t)、i fa (t)、i fb (t) Respectively the final predicted value of the load current,The predicted value of the multivariate linear regression model and the predicted value of the power dispatching system,pfor the accuracy penalty coefficients of the multiple linear regression model,qthe accuracy penalty coefficient of the power dispatching system is as follows:
wherein，i p (c)、i pa (c)、i pb (c) Respectively is a historical true value of the load current, a historical predicted value of the multiple linear regression model and a historical predicted value of the power dispatching system,cis a variable of the time, and is,nto predict the number of samples in time.
And step four, according to the original characteristic data set, considering the space distribution and time delay characteristics of the operation temperature change of the transformer, constructing a time convolution network prediction model based on multi-characteristic scale fusion, and predicting the space oil temperature and the winding temperature.
In the first step of this embodiment, the state sensing data of the power transformer includes historical data such as oil level temperature, three-phase winding temperature, load current, bias current, three-phase current unbalance, ambient temperature, ambient wind, heat sink temperature difference deviation, cooling power, vibration, oil chromatography, and the like; according to wavelet multi-resolution analysis, symlet wavelets are selected as wavelet bases, 5-layer wavelet decomposition is carried out on each dimension of historical data, threshold quantization is carried out on obtained wavelet decomposition coefficients, an extreme value threshold estimation method is adopted for threshold selection, wavelet decomposition coefficients are processed by a soft threshold denoising method, finally wavelet inverse transformation reconstruction is carried out by combining 5-layer low-frequency coefficients and all layers of high-frequency coefficients which are processed through threshold quantization, and denoised state perception data are obtained.
In the second step of this embodiment, a CART decision tree is used to select optimal feature factors: step S1: each sample in the state perception data sample set comprises data of each characteristic quantity manually selected in advance and a corresponding measured temperature value; calculating a kini coefficient of each feature according to sample data, selecting the feature with the minimum kini coefficient and a value corresponding to the feature as an optimal feature and an optimal segmentation point, and dividing a sample set into two subsets; step S2: repeating the step S1, recursively constructing a binary decision tree, and the step S3: and calculating the importance index of each feature according to the finally generated decision tree, eliminating the features with low correlation, completing feature factor screening, and constructing hot spot temperature transient curves of different structural members to predict an original feature data set.
In the third step of this embodiment, because the magnitude of the load current is a direct factor affecting the operating temperature of the transformer, and because of the uncertainty of the load current, the future short-term load current flowing through the transformer winding needs to be predicted according to the influence factor and the change rule of the historical load current. Selecting historical load current, holiday variables, air temperature variables and time variables as characteristic quantitiesxThe multiple linear regression prediction model for the future short-term load current is as follows:
whereinβIn the form of a vector of regression coefficients,ξare random error terms, and are model parameters obtained by learning,x 1 in order to be able to history the load current,x 2 e {0,1} is a holiday variable (Cx 2 =0 representing a non-holiday),x 3 for the air temperature variable, consider that the warmer or colder the air temperature, the electricity consumption will be increased, order
In which
The number of the actual air temperature degrees is,x 4 as a time variable, a sampling period is made every ten minutes.β 0 A regression coefficient representing the characteristic amount is calculated,β 1 a regression coefficient representing the historical load current,β 2 a regression coefficient representing a holiday variable,β 3 a regression coefficient representing a variable of the air temperature,β 4 regression coefficient representing time variable, error of squareThe difference, as a model error, is expressed as:
wherein the content of the first and second substances,J(β)in order to be a model error, the model error,mthe number of samples input during the training is,iis a sample number, and is a sample number,h β (x i() )in order to predict the load current (i.e.,x i() indicating a sample set with a sequence number ofiThe characteristic quantity of (a) is calculated,y i() indicating a sample set of sequence numbers ofiTraining and reducing the model error by using a gradient descent method, wherein the model parameter updating expression is as follows:
wherein the content of the first and second substances,
the regression coefficient is updated for the jth,β j is the jth of the regression coefficients for the j,αto update the step size. And ending the training until the model error is not reduced any more. Only the characteristic quantity is input when prediction is performedxThe predicted load current can be obtained
. Meanwhile, in order to improve the accuracy of load current prediction, a short-term load prediction curve of the power dispatching system is called, the prediction accuracy of two load current prediction methods, namely the short-term load prediction curve of the power dispatching system and a multiple linear regression prediction model, is tracked in real time, the final prediction current value is determined according to the accuracy penalty coefficients of the two prediction modes in the last prediction period through dynamic calculation, and the accuracy of the prediction method is dynamically improved.
In the fourth step of this embodiment, in consideration of the spatial distribution characteristic of the power transformer, six groups of spatial partitions are set, which are respectively the top layer oil temperature, the bottom layer oil temperature, the oilpaper temperature, the high-voltage winding temperature, the medium-voltage winding temperature, and the low-voltage winding temperature, that is, the data dimension of the power transformer temperature distribution and the data dimension of the time convolution network prediction model output based on the multi-feature scale fusion are 5.
Referring to fig. 2, the time convolution network prediction model construction process based on multi-feature scale fusion is as follows:
step 4.1: constructing a feature data set reconstruction module, applying different time delay parameters for each space partition in consideration of different time delay characteristics of temperature changes of different parts of the power transformer, and splicing the current time original feature data set data and the historical original feature data set data to form reconstruction feature data sets with different time scales, wherein as shown in fig. 3, the feature dimensions are described as follows:
whereind new Is the dimension of the reconstructed feature data set,d in is the dimension of the input original feature data set,d out is the dimension of the model output data,ωis a delay parameter. If the time delay parameter is 2, the reconstructed feature data set comprises original feature data of the current time and the previous two times and output data of the previous two times.
Step 4.2: and constructing a feature extraction module, and extracting a space-time feature vector according to a reconstructed feature data set by adopting a residual expansion Causal convolution unit (RDCCU) (shown in figure 4), wherein the RDCCU is formed by connecting two forward propagation modules through a residual, and each forward propagation module is formed by connecting a expanded Causal convolution layer (scaled practical Conv), a weight normalization layer (Weightnorm), a correction linear unit layer (ReLU) and a random deactivation layer (Dropout) in sequence. The dilation causal convolution layer is a core structure for extracting features, is a combination of causal convolution and dilation convolution, and can extract historical information well by only convolving output of the dilation causal convolution layer with current and past neurons of an upper layer. The weight normalization layer performs decoupling processing on the weight vector in the Euclidean norm and the direction to accelerate the network training speed. The randomly inactive layer prevents the trained overfitting by suppressing part of the neurons. The layer of corrective linear units (ReLU) is a commonly used activation function that enables neural networks to perform non-linear mapping. Respectively inputting the reconstruction characteristic data set corresponding to each space partition into an RDCCU unit to obtain corresponding space-time characteristic vectors;
step 4.3: and constructing a feature fusion module, performing weighted fusion on the space-time feature vectors output by each RDCCU unit, automatically adjusting the weight of each residual expansion causal convolution unit by adopting a self-adaptive mechanism, obtaining an initial weight by using a one-dimensional residual expansion causal convolution unit, and selecting a softmax function as an activation function of the initial weight.
Step 4.4: and inputting the fused data into a multi-layer perceptron MLP prediction module for prediction, and finally outputting predicted temperature distribution data. The model training comprises the training of a feature extraction module, a feature fusion module and a multi-layer perceptron MLP prediction module. And automatically updating network parameters by adopting an Adam optimization algorithm, and stopping training until the prediction accuracy tends to be stable.
Claims (10)
1. A power transformer short-term operation temperature prediction method based on multivariate perception is characterized by comprising the following steps:
acquiring historical state sensing data of a power transformer, preprocessing the state sensing data by using wavelet transformation, and removing noise signals in the state sensing data;
step two, performing feature screening based on the state perception data; firstly, manually selecting characteristic factors influencing the oil level and the hot spot temperature of the transformer; then, further screening out characteristic factors by using a decision tree, removing redundant data, and constructing an original characteristic data set for predicting the hotspot temperature transient curve;
constructing a multiple linear regression prediction model for the future short-term load current, and predicting the future short-term load current based on the historical load current, the holiday variables and the air temperature variables; meanwhile, calling a short-term load prediction curve of the power dispatching system, and determining a final prediction value of the load current by integrating a multiple linear regression prediction model and the prediction value of the power dispatching system;
and step four, according to the original characteristic data set, considering the space distribution and time delay characteristics of the operation temperature change of the transformer, constructing a time convolution network prediction model based on multi-characteristic scale fusion, and predicting the space oil temperature and the winding temperature.
2. The power transformer short-term operation temperature prediction method based on multivariate perception as claimed in claim 1, wherein in the first step, the state perception data of the power transformer comprises oil surface temperature, three-phase winding temperature, load current, bias current, three-phase current unbalance, ambient temperature, ambient wind, heat sink temperature difference deviation, cooling power, vibration, oil chromatography data.
3. The power transformer short-term operation temperature prediction method based on multivariate perception as claimed in claim 1, wherein in the first step, according to wavelet multi-resolution analysis, symlet wavelet is selected as a wavelet base, 5 layers of wavelet decomposition are performed on each dimension of historical data, threshold quantization is performed on the obtained wavelet decomposition coefficients, an extreme value threshold estimation method is adopted for threshold selection, the wavelet decomposition coefficients are processed by a soft threshold denoising method, and finally wavelet inverse transformation reconstruction is performed by combining the 5 th layer low-frequency coefficients and each layer high-frequency coefficients subjected to threshold quantization to obtain denoised state perception data.
4. The power transformer short-term operation temperature prediction method based on multivariate perception as claimed in claim 1, wherein in step two, the CART decision tree is used for optimal characteristic factor selection: step S1: each sample in the state perception data sample set comprises data of each characteristic quantity manually selected in advance and a corresponding measured temperature value; calculating a kini coefficient of each feature according to sample data, selecting the feature with the minimum kini coefficient and a value corresponding to the feature as an optimal feature and an optimal segmentation point, and dividing a sample set into two subsets; step S2: repeating the step S1, recursively constructing a binary decision tree, and the step S3: and calculating the importance index of each feature according to the finally generated decision tree, eliminating the features with low correlation, completing feature factor screening, and constructing hot spot temperature transient curves of different structural members to predict an original feature data set.
5. The method for predicting the short-term operation temperature of the power transformer based on multivariate perception as claimed in claim 1, wherein in the third step, the process of determining the final predicted value of the load current is as follows:
wherein，tAs a matter of time, the time is,i f (t)、i fa (t)、i fb (t) Respectively as the final predicted value of the load current, the predicted value of the multiple linear regression model and the predicted value of the power dispatching system,pfor the accuracy penalty coefficients of the multiple linear regression model,qthe accuracy penalty coefficient of the power dispatching system is as follows:
wherein，i p (c)、i pa (c)、i pb (c) Respectively is a historical true value of the load current, a historical predicted value of the multiple linear regression model and a historical predicted value of the power dispatching system,cis a variable of the time, and is,nto predict the number of samples in time.
6. The method for predicting the short-term operation temperature of the power transformer based on the multivariate perception as claimed in claim 5, wherein in the third step, the historical load current, the variables of the holidays and the festivals, the air temperature variable and the time variable are selected as characteristic quantitiesxThe multiple linear regression prediction model for the future short-term load current is as follows:
whereinβIn the form of a vector of regression coefficients,ξare random error terms, and are model parameters obtained by learning,x 1 in order to be able to history the load current,x 2 e {0,1} is a holiday variable, x 3 is a variable of the air temperature, and is,x 4 is a variable of the time, and is,β 0 a regression coefficient that represents the amount of feature,β 1 a regression coefficient representing the historical load current,β 2 a regression coefficient representing a holiday variable,β 3 a regression coefficient representing a variable of the air temperature,β 4 regression coefficients representing time variables, let the squared error be the model error, expressed as:
wherein the content of the first and second substances,J(β)in order to be a model error,mthe number of samples input for training is the number of samples,iis a sample number of a sample to be sampled,h β (x i() )in order to predict the load current(s),x i() indicating a sample set of sequence numbers ofiThe characteristic quantity of (a) is calculated,y i() indicating a sample set with a sequence number ofiTraining and reducing the model error by using a gradient descent method, wherein the model parameter updating expression is as follows:
wherein the content of the first and second substances,
the regression coefficient is updated for the jth,β j is the (j) th regression coefficient,αto update the step size.
7. The power transformer short-term operation temperature prediction method based on multivariate perception as claimed in claim 1, wherein the time convolution network prediction model based on multi-feature scale fusion comprises a feature data set reconstruction module, a feature extraction module, a feature fusion module and a multi-layer perceptron MLP prediction module which are arranged in sequence.
8. The method for predicting the short-term operation temperature of the power transformer based on the multivariate perception according to claim 7, wherein the feature extraction module adopts a residual expansion causal convolution unit, the residual expansion causal convolution unit is formed by connecting two forward propagation modules through a residual, and each forward propagation module is sequentially formed by connecting an expansion causal convolution layer, a weight normalization layer, a correction linear unit layer and a random deactivation layer.
9. The method for predicting the short-term operation temperature of the power transformer based on the multivariate perception according to claim 1, wherein the time convolution network prediction model based on the multi-feature scale fusion is constructed by the following process:
step 4.1: constructing a characteristic data set reconstruction module: splicing the current time original characteristic data set data and the historical original characteristic data set data to form a reconstructed characteristic data set with different time scales, wherein the characteristic dimensions are described as follows:
whereind new Is the dimension of the reconstructed feature data set,d in is the dimension of the input original feature data set,d out is the dimension of the output data of the model,ωis a delay parameter;
step 4.2: a feature extraction module is constructed, a residual error expansion causal convolution unit is adopted to extract space-time feature vectors according to the reconstruction feature data sets, and the reconstruction feature data sets corresponding to each space partition are respectively input into one residual error expansion causal convolution unit to obtain corresponding space-time feature vectors;
step 4.3: constructing a feature fusion module, and performing weighted fusion on the space-time feature vectors output by each residual expansion causal convolution unit;
step 4.4: and inputting the fused data into a multi-layer perceptron MLP prediction module for prediction, and finally outputting predicted temperature distribution data.
10. The method for predicting the short-term operation temperature of the power transformer based on multivariate perception as recited in claim 9, wherein in step 4.3, the weight of each residual expansion causal convolution unit is automatically adjusted by an adaptive mechanism, a one-dimensional residual expansion causal convolution unit is used to obtain an initial weight, and a softmax function is selected as an activation function of the initial weight.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211150771.XA CN115238754B (en) | 2022-09-21 | 2022-09-21 | Power transformer short-term operation temperature prediction method based on multivariate perception |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211150771.XA CN115238754B (en) | 2022-09-21 | 2022-09-21 | Power transformer short-term operation temperature prediction method based on multivariate perception |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115238754A true CN115238754A (en) | 2022-10-25 |
| CN115238754B CN115238754B (en) | 2023-02-14 |
Family
ID=83682316
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211150771.XA Active CN115238754B (en) | 2022-09-21 | 2022-09-21 | Power transformer short-term operation temperature prediction method based on multivariate perception |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115238754B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115618747A (en) * | 2022-11-23 | 2023-01-17 | 广东电网有限责任公司中山供电局 | Cable hot spot state evaluation method and device |
| CN116432406A (en) * | 2023-03-09 | 2023-07-14 | 广东电网有限责任公司佛山供电局 | Method and device for calculating hot spot temperature of working winding of oil immersed transformer |
| CN117021123A (en) * | 2023-10-09 | 2023-11-10 | 泓浒(苏州)半导体科技有限公司 | Vibration prediction system and method for wafer transfer mechanical arm in ultra-vacuum environment |
| CN117216673A (en) * | 2023-11-08 | 2023-12-12 | 国网江西省电力有限公司电力科学研究院 | Current transformer monitoring evaluation overhauls platform |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2244089A1 (en) * | 2009-04-24 | 2010-10-27 | Josef Altmann | On-line diagnostic and prediction of dielectric behavior of power transformers |
| CN108664682A (en) * | 2017-11-30 | 2018-10-16 | 全球能源互联网研究院有限公司 | A kind of prediction technique and its system of transformer top-oil temperature |
| CN110610280A (en) * | 2018-10-31 | 2019-12-24 | 山东大学 | Short-term prediction method, model, device and system for power load |
| CN111401657A (en) * | 2020-04-02 | 2020-07-10 | 国网冀北电力有限公司唐山供电公司 | Transformer hot spot temperature time sequence prediction method based on data mining algorithm |
| CN111428926A (en) * | 2020-03-23 | 2020-07-17 | 国网江苏省电力有限公司镇江供电分公司 | Regional power load prediction method considering meteorological factors |
| CN111461922A (en) * | 2020-04-02 | 2020-07-28 | 国网冀北电力有限公司唐山供电公司 | Transformer hot spot temperature real-time prediction method based on extreme learning machine |
| CN114397526A (en) * | 2022-01-14 | 2022-04-26 | 国网辽宁省电力有限公司电力科学研究院 | Power transformer fault prediction method and system driven by state holographic sensing data |
-
2022
- 2022-09-21 CN CN202211150771.XA patent/CN115238754B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2244089A1 (en) * | 2009-04-24 | 2010-10-27 | Josef Altmann | On-line diagnostic and prediction of dielectric behavior of power transformers |
| CN108664682A (en) * | 2017-11-30 | 2018-10-16 | 全球能源互联网研究院有限公司 | A kind of prediction technique and its system of transformer top-oil temperature |
| CN110610280A (en) * | 2018-10-31 | 2019-12-24 | 山东大学 | Short-term prediction method, model, device and system for power load |
| CN111428926A (en) * | 2020-03-23 | 2020-07-17 | 国网江苏省电力有限公司镇江供电分公司 | Regional power load prediction method considering meteorological factors |
| CN111401657A (en) * | 2020-04-02 | 2020-07-10 | 国网冀北电力有限公司唐山供电公司 | Transformer hot spot temperature time sequence prediction method based on data mining algorithm |
| CN111461922A (en) * | 2020-04-02 | 2020-07-28 | 国网冀北电力有限公司唐山供电公司 | Transformer hot spot temperature real-time prediction method based on extreme learning machine |
| CN114397526A (en) * | 2022-01-14 | 2022-04-26 | 国网辽宁省电力有限公司电力科学研究院 | Power transformer fault prediction method and system driven by state holographic sensing data |
Non-Patent Citations (2)
| Title |
|---|
| ALI DEIHIMI等: ""Short-term electric load and temperature forecasting using wavelet echo state networks with neural reconstruction"", 《ENERGY》 * |
| 闫优俊: ""基于多元非线性回归算法的动车组变压器温度预测研究"", 《控制与信息技术》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115618747A (en) * | 2022-11-23 | 2023-01-17 | 广东电网有限责任公司中山供电局 | Cable hot spot state evaluation method and device |
| CN115618747B (en) * | 2022-11-23 | 2023-03-21 | 广东电网有限责任公司中山供电局 | Cable hot spot state evaluation method and device |
| CN116432406A (en) * | 2023-03-09 | 2023-07-14 | 广东电网有限责任公司佛山供电局 | Method and device for calculating hot spot temperature of working winding of oil immersed transformer |
| CN116432406B (en) * | 2023-03-09 | 2024-02-02 | 广东电网有限责任公司佛山供电局 | Method and device for calculating hot spot temperature of working winding of oil immersed transformer |
| CN117021123A (en) * | 2023-10-09 | 2023-11-10 | 泓浒(苏州)半导体科技有限公司 | Vibration prediction system and method for wafer transfer mechanical arm in ultra-vacuum environment |
| CN117021123B (en) * | 2023-10-09 | 2024-01-30 | 泓浒(苏州)半导体科技有限公司 | Vibration prediction system and method for wafer transfer mechanical arm in ultra-vacuum environment |
| CN117216673A (en) * | 2023-11-08 | 2023-12-12 | 国网江西省电力有限公司电力科学研究院 | Current transformer monitoring evaluation overhauls platform |
| CN117216673B (en) * | 2023-11-08 | 2024-03-12 | 国网江西省电力有限公司电力科学研究院 | Current transformer monitoring evaluation overhauls platform |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115238754B (en) | 2023-02-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115238754B (en) | Power transformer short-term operation temperature prediction method based on multivariate perception | |
| JP7242101B2 (en) | Engine surge failure prediction system and method based on fusion neural network model | |
| CN110929847A (en) | Converter transformer fault diagnosis method based on deep convolutional neural network | |
| CN111079989B (en) | DWT-PCA-LSTM-based water supply amount prediction device for water supply company | |
| CN109472097B (en) | Fault diagnosis method for online monitoring equipment of power transmission line | |
| CN112464566A (en) | Transformer oil temperature prediction method based on genetic algorithm and BP neural network | |
| CN111380686A (en) | XGboost algorithm model-based fan main bearing fault monitoring and diagnosis method | |
| CN112598151A (en) | High-voltage cable surface temperature prediction method in cable trench laying mode | |
| CN112070322B (en) | High-voltage cable line running state prediction method based on long-short term memory network | |
| CN114283120B (en) | Domain-adaptive-based end-to-end multisource heterogeneous remote sensing image change detection method | |
| CN113343581B (en) | Transformer fault diagnosis method based on graph Markov neural network | |
| JP2023550995A (en) | Load prediction of electrical equipment using machine learning | |
| CN115293326A (en) | Training method and device of power load prediction model and power load prediction method | |
| CN114358389A (en) | Short-term power load prediction method combining VMD decomposition and time convolution network | |
| CN115618732A (en) | Nuclear reactor digital twin key parameter autonomous optimization data inversion method | |
| CN116663419A (en) | Sensorless equipment fault prediction method based on optimized Elman neural network | |
| CN115983374A (en) | Cable partial discharge database sample expansion method based on optimized SA-CACGAN | |
| CN117407675A (en) | Lightning arrester leakage current prediction method based on multi-variable reconstruction combined dynamic weight | |
| CN117290685A (en) | Power plant power equipment expert diagnosis system and method based on historical data | |
| CN113434493A (en) | Non-invasive load decomposition method based on Transformer | |
| CN117592593A (en) | Short-term power load prediction method based on improved quadratic modal decomposition and WOA optimization BILSTM-intent | |
| CN116417992B (en) | Fan power prediction model establishment method and application based on space-time attention mechanism | |
| CN117034169A (en) | Power grid main transformer equipment abnormal state prediction method based on time sequence causality network | |
| CN115689358A (en) | Power distribution network state estimation method based on data dimension reduction and related device | |
| CN115935814A (en) | Transformer fault prediction method based on ARIMA-SVM model |
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 |