CN115796336A - Transformer area line loss prediction method based on feature selection and GRU network - Google Patents
Transformer area line loss prediction method based on feature selection and GRU network Download PDFInfo
- Publication number
- CN115796336A CN115796336A CN202211387524.1A CN202211387524A CN115796336A CN 115796336 A CN115796336 A CN 115796336A CN 202211387524 A CN202211387524 A CN 202211387524A CN 115796336 A CN115796336 A CN 115796336A
- Authority
- CN
- China
- Prior art keywords
- line loss
- transformer area
- model
- data
- gate
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention provides a transformer area line loss prediction method based on feature selection and a GRU network, which comprises the following steps of firstly selecting a multi-dimensional electrical feature index to obtain the influence degree of each electrical feature on a line loss rate; then selecting an evaluation index by taking the average percentage error as a characteristic to obtain an optimal electrical characteristic dimension which is beneficial to line loss evaluation; and finally, establishing a line loss prediction model of the GRU network based on the selected optimal electrical characteristic index to obtain line loss prediction results of different transformer areas. The invention provides a transformer area line loss prediction method based on feature selection and a GRU network, which can effectively improve line loss prediction precision, and provides an effective way for analyzing transformer area line loss influence factors and transformer area line loss abnormity, thereby improving the fine management of a transformer area.
Description
Technical Field
The invention belongs to the technical field of line loss rate, and particularly relates to a station area line loss prediction method based on feature selection and a GRU network.
Background
In the operation process of a power grid, electric energy is transmitted to a terminal user by means of various electric equipment, and certain loss is generated in the operation process of the electric energy due to the fact that the electric equipment has impedance and the equipment generates errors, and the loss is called line loss. At present, the important links of line loss generation in China are mainly concentrated in 380V low-voltage power distribution networks, according to the measurement and calculation of national power grid companies, 70 hundred million KWH electric energy can be used in China as long as the line loss rate is reduced by 1%, meanwhile, each company also formulates an effective loss reduction scheme for line loss treatment, so that the line loss rate is gradually reduced, but the overall line loss rate of the power grid is still at a higher level, therefore, accurate line loss prediction of a transformer area has great research value, and the method has certain guiding significance for energy saving loss of the transformer area.
In the line loss management of a power supply enterprise, the fine management of a distribution area generally adopts the forms of partitioning, voltage division, branching and distribution area, wherein the distribution area refers to a management method for calculating, analyzing and evaluating the power loss according to the power supply area of each public distribution transformer in a power grid. However, the environment of an actual field is not considered in part of low-voltage transformer areas, so that various faults of the acquisition device cannot normally acquire transformer area data, the transformer area line loss rate cannot be actually calculated, and the transformer area line loss analysis is relatively large; the traditional low-voltage transformer area line loss calculation method depends on detailed operation parameters and a complex transformer area structure, and the traditional method has too many assumed conditions and low accuracy. In addition, massive power data generated by the development of the smart grid are not completely applied to the analysis aspect of the line loss of the transformer area.
The line loss rate plays an important role in evaluating the economic operation of the power system, so that the research on the line loss condition of the distribution room and the analysis of important electrical characteristics influencing the line loss of the distribution room have important research significance for improving the economic operation level of the power distribution network.
Disclosure of Invention
In order to solve the problems, the invention provides a station area line loss prediction method based on feature selection and a GRU network, which can accurately analyze the abnormity of the station area line loss rate and provide reasonable suggestions for stable and safe operation of a station area power system.
In view of the above-mentioned deficiencies in the prior art, the present invention provides a method for predicting line loss of a distribution room based on feature selection and a GRU network, wherein the method for predicting line loss of the distribution room comprises the following steps:
step (1): collecting multi-dimensional electrical characteristic index parameter data of the transformer area based on the power utilization management system, and constructing original data influencing the line loss rate of the transformer area, wherein the original data comprise electrical characteristic indexes and actual line loss rate data of the transformer area;
step (2): normalizing the original data set, and simultaneously detecting abnormal values of abnormal line loss of the distribution room by adopting a box type graph model to obtain standard distribution room data, and then randomly dividing the standard distribution room data into training set data and testing set data;
and (3): performing correlation analysis on the multi-dimensional electrical characteristic indexes, extracting important factors influencing the line loss rate and the optimal electrical characteristic index dimension by adopting a correlation coefficient method and an XGboost model, and constructing an optimal electrical characteristic index system of the line loss of the transformer area;
and (4): constructing a GRU neural network learning model, and setting basic parameters of the model, such as an activation function, a loss function and the like;
and (5): taking the optimal electrical characteristic index system training data set in the step (3) as input, training GRU network parameters, and obtaining a transformer area line loss prediction model;
and (6): and inputting the test data set into the trained GRU network model, and performing calculation precision analysis and input feature validity analysis on the line loss rate of the transformer area.
Further, the step (1) of obtaining the multidimensional electrical characteristic index parameters of the distribution room comprises: region type, wire impedance ratio, comprehensive multiplying power, resident capacity, daily load rate, power supply quantity, daily power factor, theoretical line loss rate, CT variable ratio value, CT value, transformer area capacity, non-resident number of households, maximum zero line current, non-resident capacity, daily reactive power quantity, daily active power quantity, three-phase electric energy meter number, lost electric quantity and single-phase electric energy meter number.
Further, the normalization processing in the step (2) adopts a linear function method, and a specific calculation formula is as follows:
wherein x max Is the maximum value of the raw data, x mi n is the minimum value of the original data, x is the original data, x * The result is after eliminating the dimension influence.
Further, the step (3) specifically comprises the steps of primarily selecting electrical characteristics based on the Spearman correlation coefficient and further selecting an optimal electrical characteristic index based on the XGboost model;
the specific method for preliminarily selecting the electrical characteristics based on the Spearman correlation coefficient comprises the following steps of:
let the total number of samples be m, R i And S i Rank the ith number of two random variables;
and
is the average level of two random variables, then twoThe correlation coefficient values between random variables are:
the specific method for further selecting the optimal electrical characteristic index based on the XGboost model comprises the following steps:
(1) Determining a regression tree model based on CART to establish an XGboost model as follows:
in the formula: m is the original total sample number; Γ is the collection space of all trees;
(2) Constructing an original objective function of the XGboost model and adding a regularization term as follows:
wherein, L (y) i ,F t (x i ) Is the training error between the predicted value and the actual value; omega (f) i ) Is a regularization term, and the specific calculation formula is as follows:
wherein T indicates that the tree has T leaf nodes, w j Representing the weight of the jth leaf node, wherein gamma is a regular coefficient, and lambda is a difficulty coefficient of tree segmentation;
(3) And (3) carrying out second-order Taylor expansion derivation on the target function to obtain the optimal solution of the target function as follows:
wherein, in the formula: w is a j * Is the jth leaf nodeThe optimal score value of the point; g j First derivative on the loss function for all data; h j Second derivatives of all data on the loss function;
(4) And substituting the optimal solution into the original objective function to obtain:
further, the building of the GRU line loss prediction model in the step (4) includes the following basic principles and calculation steps:
(1) The GRU neural network has strong learning ability and function fitting ability, is an optimization improvement of the traditional LSTM network, simplifies and combines a forgetting gate and an input gate of the LSTM into a single updating gate, and uses a hidden state to transmit information. The reset gate is used for controlling the state information of the previous time step hidden layer to be transferred to the current time step candidate hidden state so as to reset the current time step candidate hidden state information, and is the combination of the memory unit and the hidden layer. The updating gate is used for controlling the previous time step hidden state information to be transferred to the current time step hidden state, and then updating the current time step hidden state information, and is the combination of the forgetting gate and the input gate. The GRU network reduces the parameters of the network model by simplifying the structure of the door, accelerates the training speed of the model and improves the performance of the network model.
(2) LSTM includes 4 parts per neural unit, namely input gate i t Forgetting door f t Output gate O t And a memory cell C t (ii) a The specific calculation steps are as follows:
input state
g t ＝tanh(W ig x t +b ig +W hc h t-1 +b hg )，
Gated state
i t ＝sigmoid(W ij x t +b ii +W hi h t-1 +b hi )，
f t ＝sigmoid(W if x t +b if +W hf h t-1 +b hf )，
O t ＝sigmoid(W io x t +b io +W ho h t-1 +b ho )
Memory state
C t ＝f t ×C t-1 +i t ×g t ，
Output state
h t ＝o t ×tanh(C t )，
Wherein tanh (·) is a hyperbolic tangent function, sigmoid (·) is a sigmoid function, W is a gate weight vector, and b is a bias phase;
the input gate, the forgetting gate and the output gate of the LSTM are changed into an updating gate z t And a reset gate r t Combining the unit states c and h into a state h to obtain a GRU network learning model,
the specific calculation steps are as follows:
z t ＝sigmoid(W z X t +U z H t-1 +b z )，
r t ＝sigmoid(W r X t +U r H t-1 +b r )，
further, an Adam algorithm is selected on the optimization algorithm of the GRU network learning model in the step (6), a Dropout layer discarding rate is set to be 0.2, and a loss function is a Huber loss function.
Compared with the prior art, the beneficial effects are:
1. according to the method, the extraction of key electrical characteristic indexes affecting the line loss rate is realized by combining a Spearman correlation coefficient method and an XGboost model on the basis of actual operation data of the conventional power utilization acquisition system and the like, the electrical characteristic dimension of a transformer area is reduced, the model complexity is reduced, and the operation time is shortened.
2. The method adopts a GRU network model in deep learning, and has the advantages of simple network structure, less parameters to be optimized, flexible kernel function selection, strong robustness, strong nonlinear fitting capability and the like.
3. The invention realizes the prediction of the line loss of the transformer area and the extraction of important electrical indexes influencing the line loss rate, provides help for the lean management of the transformer area and defines the loss reduction direction.
Drawings
Fig. 1 is a flowchart of the operation of a method for predicting line loss of a distribution room based on feature selection and a GRU network according to the present invention;
FIG. 2 is an electrical characteristic index plot based on Spearman correlation coefficient in an embodiment of the present invention
FIG. 3 is a graph of the electrical feature importance score ranking based on the XGboost model in an embodiment of the invention;
FIG. 4 is a graph of the relationship between the number of electrical features and MAPE in accordance with one embodiment of the present invention;
fig. 5 is a unit structure of a GRU network according to an embodiment of the present invention;
FIG. 6 is a diagram illustrating predicted results of a portion of a testbench area in accordance with an embodiment of the present invention.
Detailed Description
The following describes in detail a specific embodiment of the method for predicting the line loss of the distribution room based on the feature selection and the GRU network according to the present invention with reference to the accompanying drawings.
The invention provides a station area line loss prediction method based on feature selection and a GRU network. On one hand, a Speramman correlation coefficient method and an XGboost algorithm are combined to screen the multi-dimensional electrical characteristic indexes of the transformer area, so that the complexity of data is reduced, and the calculation time is shortened; on the other hand, the selected important electrical characteristic indexes are used as the input of the GRU network model, and the calculation accuracy of the line loss of the transformer area is further improved.
As shown in fig. 1, the method for predicting the line loss of the distribution room of the present invention includes the following steps:
step 1: collecting multi-dimensional electrical characteristic index parameter data of the transformer area based on the power utilization management system, and constructing original data which influence the line loss rate of the transformer area, wherein the original data comprise electrical characteristic indexes and actual line loss rate data of the transformer area;
and 2, step: normalizing the original data set, and simultaneously detecting abnormal values of abnormal line loss of the distribution room by adopting a box-type graph model to obtain standard distribution room data, and then randomly dividing the standard distribution room data into training set data and test set data;
and step 3: performing correlation analysis on the multi-dimensional electrical characteristic indexes, extracting important factors influencing the line loss rate by adopting a correlation coefficient method and an XGboost model, and constructing an optimal electrical characteristic index system of the line loss of the transformer area;
and 4, step 4: building a GRU neural network learning model, and setting basic parameters of the model, such as an activation function, a loss function and the like;
and 5: taking the optimal electrical characteristic index system training data set in the step 3 as input, training GRU network parameters, and obtaining a transformer area line loss prediction model;
and 6: and inputting the test data set into the trained GRU network model, and performing calculation precision analysis and input feature validity analysis on the transformer area line loss rate.
In this embodiment, the multidimensional electrical characteristic index parameter of the distribution room includes: region type, wire impedance ratio, comprehensive multiplying power, resident capacity, daily load rate, power supply quantity, daily power factor, theoretical line loss rate, CT variable ratio value, CT value, transformer area capacity, non-resident number of households, maximum zero line current, non-resident capacity, daily reactive power quantity, daily active power quantity, three-phase electric energy meter number, lost electric quantity and single-phase electric energy meter number.
In this embodiment, the data normalization processing adopts a linear function programming processing technique, which is specifically as follows:
let x max Is the maximum value of the raw data, x min Is the minimum value of the original data, x is the original data, x * In order to eliminate the effect of the dimension as a result,
in the embodiment, the electrical characteristic index is extracted by combining a Spearman correlation coefficient and an XGBoost algorithm, which are specifically as follows:
the specific method for preliminarily selecting the electrical characteristics based on the Spearman correlation coefficient comprises the following steps of:
let the total number of samples be m, R i And S i Rank the ith number of two random variables;
and
the average level of the two random variables is the correlation coefficient value between the electrical index and the line loss rate:
screening out an electrical index with a Spearman correlation coefficient absolute value larger than 0.1 as a primary screening result;
the specific method for further selecting the optimal electrical characteristic index based on the XGboost model comprises the following steps:
(1) The XGboost model is established based on the regression tree model of CART and comprises the following steps:
(2) Constructing an original objective function of the XGboost model and adding a regularization term as follows:
wherein, L (y) i ,F t (x i ) Is a training error between the predicted value and the actual value；Ω(f i ) Is a regularization term, and the specific calculation formula is as follows:
wherein T indicates that the tree has T leaf nodes, w j Representing the weight of the jth leaf node, wherein gamma is a regular coefficient, and lambda is a difficulty coefficient of tree segmentation;
(3) And (3) carrying out second-order Taylor expansion derivation on the target function to obtain the optimal solution of the target function as follows:
in the formula: w is a j * The optimal score value of the jth leaf node; g j First derivative on the loss function for all data; h j Second derivatives of all data on the loss function;
(4) And substituting the optimal solution into an original objective function to obtain:
in the present embodiment, a GRU neural network model is adopted, the network structure of which is shown in fig. 5,
the specific calculation steps of the GRU neural network are as follows:
z t ＝sigmoid(W z X t +U z H t-1 +b z )，
r t ＝sigmoid(W r X t +U r H t-1 +b r )，
example 1
In this embodiment, the experimental data is composed of electrical characteristic parameters of 117028 transformer areas collected in a certain city of east china, including 20 electrical characteristic indexes closely related to the line loss rate, that is: region type, wire impedance ratio, comprehensive multiplying power, resident capacity, daily load rate, power supply quantity, daily power factor, theoretical line loss rate, CT variable ratio value, CT value, transformer area capacity, non-resident number of households, maximum zero line current, non-resident capacity, daily reactive power quantity, daily active power quantity, three-phase electric energy meter number, lost electric quantity and single-phase electric energy meter number.
Screening is carried out according to the provided electrical characteristic indexes, a Spearman correlation coefficient value of the electrical characteristic and an importance ranking of the electrical characteristic are respectively shown in fig. 2 and fig. 3, a change curve of a calculation error of a model along with the electrical characteristic indexes is shown in fig. 4, it is easy to see that when the electrical characteristic dimension is 9, the calculation error of the model is the minimum, and at the moment, the feature of 9 before the electrical characteristic score ranking is selected to be beneficial to improving the prediction accuracy of the model.
And (3) inputting 90% of the station area data of the total sample as a station area line loss model of the GRU network learning model, and verifying and comparing the actual data of the station area of the last 10% with the prediction result of the method provided by the patent.
In this example, the results were analyzed using the following three evaluation indexes:
1) Mean Absolute Error (MAE)
y pre Representing the estimated line loss value, y i Representing the actual line loss value of the transformer area, wherein n is a data sample; the average absolute error is the average value of absolute errors between the actual line loss value and the estimated line loss value of the transformer area, and the larger the prediction error is in actual analysis, the larger the average absolute error is, and the smaller the opposite average absolute error is;
2) Root Mean Square Error (RMSE)
y pre Representing the estimated line loss value, y i Representing the actual line loss value of the transformer area, wherein n is a data sample; the root mean square error is essentially consistent with the mean square error, and the root-opening calculation is carried out on the basis of the mean square error, so that the original magnitude of data is well reserved, and the measurement precision is well reflected;
3) Determinable coefficient (R2)
y pre Representing the estimated line loss value, y i Representing the actual line loss value of the transformer area, n is the total number of the transformer areas,
representing the average value of the actual line loss values of the transformer area; the determinable coefficients are typically used to analyze the effective fit of the model, and normally have values in the range of [0,1 ]]The closer the value is to 1, the stronger the prediction ability of the model is, and the higher the accuracy is.
Definition E a The calculated error statistics for the GRU network model are given in table 1 as the relative error percentage of the line loss evaluation results for the transformer area. It can be seen that, in the calculation result of the GRU model, the station data with the relative error smaller than 0.5% accounts for 98.5% of the entire test station, and the number of the station data with the relative error larger than 0.5% is only 154, so that it can be shown that the line loss rate obtained by the GRU line loss evaluation model provided herein is closer to an actual value, and has higher calculation accuracy.
TABLE 1 GRU network model calculation error results
In order to further evaluate the accuracy of the GRU network model on the prediction result of the transformer substation area line loss, the prediction result is compared with the prediction results of the GBDT, MLP and BP prediction algorithms. The evaluation results are shown in table 2.
TABLE 2 evaluation results of indices of the respective prediction models
As can be seen from the results of various model indexes in the table 2, the results obtained by the method are superior to other calculation models in both the MAE error index and the RMSE error index, the coefficient determining value is as high as 0.942, and the fitting effect of the model is shown to be best, so that the method has higher accuracy in line loss prediction, higher generalization capability and higher use value. In conclusion, the method and the device can realize the prediction of the line loss rate of the transformer area and the analysis of the electrical characteristic indexes influencing the line loss rate, and can be used for practical engineering application.
As illustrated in fig. 6, is the prediction result of a part of the test station area.
Finally, it should be noted that the above embodiments are only used for illustrating the technical solutions of the present invention and not for limiting the same. It will be understood by those skilled in the art that various modifications and equivalents may be made to the embodiments of the invention as described herein, and such modifications and variations are intended to be within the scope of the claims appended hereto.
Claims (6)
1. A transformer area line loss prediction method based on feature selection and GRU network is characterized by comprising the following steps:
step (1): collecting multi-dimensional electrical characteristic index parameter data of the transformer area based on the power utilization management system, and constructing original data which influence the line loss rate of the transformer area, wherein the original data comprise electrical characteristic indexes and actual line loss rate data of the transformer area;
step (2): normalizing the original data set, and simultaneously detecting abnormal values of abnormal line loss of the distribution room by adopting a box-type graph model to obtain standard distribution room data, and then randomly dividing the standard distribution room data into training set data and test set data;
and (3): performing correlation analysis on the multidimensional electrical characteristic indexes, extracting important factors influencing the line loss rate and the optimal electrical characteristic index dimension by adopting a correlation coefficient method and an XGboost model, and constructing an optimal electrical characteristic index system of the line loss of the transformer area;
and (4): building a GRU neural network learning model, and setting basic parameters of the model, such as an activation function, a loss function and the like;
and (5): taking the optimal electrical characteristic index system training data set in the step (3) as input, training GRU network parameters, and obtaining a transformer area line loss prediction model;
and (6): and inputting the test data set into the trained GRU network model, and performing calculation error analysis on the platform area line loss rate of the test sample.
2. The method according to claim 1, wherein the method for predicting the line loss of the transformer area based on feature selection and a GRU network is characterized in that in the step (1), the multidimensional electrical feature index parameter of the transformer area comprises: region type, wire impedance ratio, comprehensive multiplying power, resident capacity, daily load rate, power supply quantity, daily power factor, theoretical line loss rate, CT variable ratio value, CT value, transformer area capacity, non-resident number of households, maximum zero line current, non-resident capacity, daily reactive power quantity, daily active power quantity, three-phase electric energy meter number, lost electric quantity and single-phase electric energy meter number.
3. The method for predicting the line loss of the transformer area based on the feature selection and the GRU network as claimed in claim 1, wherein the normalization processing in the step (2) adopts a linear function method, and a specific calculation formula is as follows:
wherein x is max Is the maximum value of the raw data, x min Is the minimum value of the original data, x is the original data, x * For eliminating knots after dimensional influenceAnd (5) fruit.
4. The feature selection and GRU network-based distribution room line loss prediction method according to claim 1, wherein the step (3) specifically comprises preliminarily selecting electrical features based on a Spearman correlation coefficient and further selecting an optimal electrical feature index based on an XGboost model;
the specific method for preliminarily selecting the electrical characteristics based on the Spearman correlation coefficient comprises the following steps of:
let the total number of samples be m, R i And S i The level of the ith value is two random variables;
and
the average level of the two random variables is, the correlation coefficient value between the two random variables is:
the specific method for further selecting the optimal electrical characteristic index based on the XGboost model comprises the following steps:
(1) The XGboost model generally adopts a regression tree model based on CART, and comprises the following steps:
in the formula: m is the original total sample number; Γ is the aggregate space of all trees, each f (x) corresponds to an independent tree and leaf weight w q (x)；
(2) Constructing an original objective function of the XGboost model and adding a regularization term as follows:
wherein, L (y) i ,F t (x i ) Is the training error between the predicted value and the actual value; omega (f) i ) Is a regularization term, and the specific calculation formula is as follows:
wherein T indicates that the tree has T leaf nodes, w j Representing the weight of the jth leaf node, wherein gamma is a regular coefficient, and lambda is a difficulty coefficient of tree segmentation;
(3) And performing second-order Taylor expansion on the target function, and then solving a partial derivative to obtain the optimal solution of the target function as follows:
in the formula: w is a j * The optimal score value of the jth leaf node; g j First derivative on the loss function for all data; h j Second derivative of all data on the loss function;
(4) And substituting the optimal solution into the original objective function to obtain:
5. the feature selection and GRU network-based transformer district line loss prediction method according to claim 1, wherein a GRU line loss prediction model is established in step (4), and the basic principle and the calculation steps are as follows:
(1) The GRU neural network has strong learning ability and function fitting ability, is an optimization improvement of the traditional LSTM network, simplifies and combines a forgetting gate and an input gate of the LSTM into a single updating gate, and uses a hidden state to transmit information; the reset gate is used for controlling the state information of the previous time step hidden layer to be transferred to the current time step candidate hidden state so as to reset the current time step candidate hidden state information, and is the combination of a memory unit and a hidden layer; the updating gate is used for controlling the previous time step hidden state information to be transferred to the current time step hidden state, so as to update the current time step hidden state information, and is the combination of a forgetting gate and an input gate; the GRU network reduces the parameters of a network model by simplifying a door structure, accelerates the training speed of the model and improves the performance of the network model;
(2) LSTM includes 4 parts per neural unit, namely input gate i t Forgetting door f t Output gate O t And memory cell C t (ii) a The specific calculation steps are as follows:
input state
g t ＝tanh(W ig x t +b ig +W hc h t-1 +b hg )，
Gated state
i t ＝sigmoid(W ij x t +b ii +W hi h t-1 +b hi )，
f t ＝sigmoid(W if x t +b if +W hf h t-1 +b hf )，
O t ＝sigmoid(W io x t +b io +W ho h t-1 +b ho )，
Memory state
C t ＝f t ×C t-1 +i t ×g t ，
Output state
h t ＝o t ×tanh(C t )，
Wherein tanh (·) is a hyperbolic tangent function, sigmoid (·) is a sigmoid function, W is a gate weight vector, and b is a bias phase;
the input gate, the forgetting gate and the output gate of the LSTM are changed into an updating gate z t And a reset gate r t In the form of a unitCombining the states c and h into a state h to obtain a GRU network learning model,
the specific calculation steps are as follows:
z t ＝sigmoid(W z X t +U z H t-1 +b z )，
r t ＝sigmoid(W r X t +U r H t-1 +b r )，
6. the method for predicting the line loss of the transformer area based on the feature selection and the GRU network as claimed in claim 1, wherein an Adam algorithm is selected on the optimization algorithm of the learning model of the GRU network in the step (6), a Dropout layer discarding rate is set to 0.2, and a loss function is a Huber loss function.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211387524.1A CN115796336A (en) | 2022-11-07 | 2022-11-07 | Transformer area line loss prediction method based on feature selection and GRU network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211387524.1A CN115796336A (en) | 2022-11-07 | 2022-11-07 | Transformer area line loss prediction method based on feature selection and GRU network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115796336A true CN115796336A (en) | 2023-03-14 |
Family
ID=85435970
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211387524.1A Pending CN115796336A (en) | 2022-11-07 | 2022-11-07 | Transformer area line loss prediction method based on feature selection and GRU network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115796336A (en) |
-
2022
- 2022-11-07 CN CN202211387524.1A patent/CN115796336A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109546659B (en) | Power distribution network reactive power optimization method based on random matrix and intelligent scene matching | |
| CN113064093B (en) | Method and system for jointly estimating state of charge and state of health of energy storage battery | |
| CN112990556A (en) | User power consumption prediction method based on Prophet-LSTM model | |
| CN108155648A (en) | Method for estimating state based on the infinite Extended Kalman filter of adaptive H | |
| CN112990500B (en) | Transformer area line loss analysis method and system based on improved weighted gray correlation analysis | |
| CN107565553A (en) | A kind of power distribution network robust dynamic state estimator method based on UKF | |
| CN110598854A (en) | GRU model-based transformer area line loss rate prediction method | |
| CN110751318A (en) | IPSO-LSTM-based ultra-short-term power load prediction method | |
| CN112149873A (en) | Low-voltage transformer area line loss reasonable interval prediction method based on deep learning | |
| CN110866366A (en) | XGboost algorithm-based island detection method for photovoltaic microgrid containing PHEV | |
| CN111525587B (en) | Reactive load situation-based power grid reactive voltage control method and system | |
| CN112269134A (en) | Battery SOC and SOH joint estimation method based on deep learning | |
| CN110837915A (en) | Low-voltage load point prediction and probability prediction method for power system based on hybrid integrated deep learning | |
| CN106296434A (en) | A kind of Grain Crop Yield Prediction method based on PSO LSSVM algorithm | |
| CN116796403A (en) | Building energy saving method based on comprehensive energy consumption prediction of commercial building | |
| CN114357670A (en) | Power distribution network power consumption data abnormity early warning method based on BLS and self-encoder | |
| CN113379116A (en) | Cluster and convolutional neural network-based line loss prediction method for transformer area | |
| CN113033898A (en) | Electrical load prediction method and system based on K-means clustering and BI-LSTM neural network | |
| CN112446175A (en) | Distributed power supply site selection optimization method based on power distribution network probability load flow calculation | |
| CN112508254A (en) | Method for determining investment prediction data of transformer substation engineering project | |
| CN116561569A (en) | Industrial power load identification method based on EO feature selection and AdaBoost algorithm | |
| CN115796336A (en) | Transformer area line loss prediction method based on feature selection and GRU network | |
| CN115330085A (en) | Wind speed prediction method based on deep neural network and without future information leakage | |
| CN113361187A (en) | Power grid line operation safety optimization method, system and storage medium | |
| Shi et al. | Study on stability feature extraction of power system using deep learning |
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 |