CN114545147A - Voltage sag source positioning method based on deep learning in consideration of time-varying topology - Google Patents

Voltage sag source positioning method based on deep learning in consideration of time-varying topology Download PDF

Info

Publication number
CN114545147A
CN114545147A CN202210025828.7A CN202210025828A CN114545147A CN 114545147 A CN114545147 A CN 114545147A CN 202210025828 A CN202210025828 A CN 202210025828A CN 114545147 A CN114545147 A CN 114545147A
Authority
CN
China
Prior art keywords
data
model
sample data
training
voltage
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
Application number
CN202210025828.7A
Other languages
Chinese (zh)
Inventor
邓亚平
贾颢
同向前
王璐
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Xian University of Technology
Original Assignee
Xian University of Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Xian University of Technology filed Critical Xian University of Technology
Priority to CN202210025828.7A priority Critical patent/CN114545147A/en
Publication of CN114545147A publication Critical patent/CN114545147A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
    • G01R31/08—Locating faults in cables, transmission lines, or networks
    • G01R31/081—Locating faults in cables, transmission lines, or networks according to type of conductors
    • G01R31/086—Locating faults in cables, transmission lines, or networks according to type of conductors in power transmission or distribution networks, i.e. with interconnected conductors
    • 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
    • G06N3/045—Combinations of networks
    • 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
    • G06N3/048—Activation functions
    • 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
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2113/00—Details relating to the application field
    • G06F2113/04—Power grid distribution networks
    • 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)
  • Theoretical Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • General Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Mathematical Physics (AREA)
  • Computational Linguistics (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • Supply And Distribution Of Alternating Current (AREA)

Abstract

The invention discloses a voltage sag source positioning method based on deep learning in time-varying topology, which solves the problems of unavailable model, poor adaptability, low positioning accuracy and poor algorithm generalization performance of the existing method when facing a time-varying power grid topological structure. By combining a deep learning model based on 1D convolution and migration learning, three-phase monitoring voltages corresponding to a limited number of monitoring nodes collected by the power quality monitoring equipment are analyzed, and then a specific line where a voltage sag source is located in the power distribution network is obtained. The positioning accuracy of the voltage sag source of the model during topology change is improved, and the self-adaptive capacity and the generalization performance of the model are improved.

Description

Voltage sag source positioning method based on deep learning in consideration of time-varying topology
Technical Field
The invention belongs to the technical field of power quality analysis, and relates to a voltage sag source positioning method based on deep learning in consideration of time-varying topology.
Background
Voltage sag affects widely relative to other power quality issues. In high-end production and manufacturing industries such as microelectronics, semiconductors, biomedical optoelectronics, aerospace and aviation technologies and the like, the voltage sag can cause product scrapping on a production line and even possibly cause equipment damage, thereby bringing about huge economic loss.
Accurately tracing the voltage sag source has important significance for dividing responsibility and accelerating the electric power marketization process. However, most of the existing data-driven voltage sag source positioning methods are to position the voltage sag source when the grid topology change is not considered, that is, the influence of the grid topology change on the voltage sag source positioning result is rarely considered, and the grid topology is not used as the input of the voltage sag source positioning analysis model. However, in the actual power grid operation process, the system topology is time-varying, and the historical database is difficult to contain massive historical data of all topologies to make up for the problem that new topology samples are few. Therefore, when the existing voltage sag source positioning method faces a time-varying power grid topological structure, the electrical characteristic quantity after the voltage sag occurs changes significantly, so that the existing method is difficult to realize accurate voltage sag source positioning.
In summary, the following difficult problems still exist in the current voltage sag source positioning method based on data driving, and need to be solved:
1. the influence of the power grid topology change on the voltage sag source positioning result is not considered, so that the data driving voltage sag source positioning method is not available.
2. The topological change of the power grid is not used as the input of a data driving model, and the model has poor adaptivity and generalization performance.
Disclosure of Invention
The invention aims to provide a voltage sag source positioning method based on deep learning in consideration of time-varying topology, and solves the problems that a model is unavailable, adaptability is poor, positioning accuracy is low and algorithm generalization performance is poor in the case of a time-varying power grid topological structure in the conventional method.
The technical scheme adopted by the invention is that the voltage sag source positioning method based on deep learning in the time-varying topology is considered, and the method specifically comprises the following steps:
step 1, collecting bus voltage data under different operating conditions, and taking the sample data as sample data corresponding to a basic power grid topological structure;
step 2, calculating the voltage root mean square value of the sample data acquired in the step 1, and if the voltage root mean square value is reduced to 90% -10% of a rated value, forming a group of voltage sag data samples by the sample data corresponding to the voltage root mean square value; otherwise, kicking off the sample data from the sample data acquired in the step 1 to obtain the preprocessed monitoring voltage data;
step 3, carrying out sample marking on the data obtained in the step 2, wherein the input data of each sample is the three-phase voltage amplitude data of the monitoring node preprocessed in the step 2, and the output data is a line sequence where a specific voltage sag source is located in the power distribution network;
step 4, randomly dividing the sample data marked in the step 3, wherein 80% of the sample data is used as a training sample data set, and the rest 20% is used as a test sample data set;
step 5, building a 1D convolution deep learning model as a basic model;
step 6, training the model set up in the step 5 by using the training sample data set in the step 4, updating parameters by using a back propagation algorithm, training by using an Adam optimizer, and obtaining the optimal neural network parameters of the trained basic model, wherein the loss function is a cross soil moisture loss function;
step 7, testing the model obtained after the training in the step 6 by using the test sample data set in the step 4, and if an overfitting phenomenon does not occur, taking the model as a final basic model; otherwise, testing the model obtained after the training in the step 6 by using the test sample data set in the step 4 again until no overfitting phenomenon occurs;
step 8, comparing with the basic power grid topology in the step 1, once one or more branches are changed due to faults in the operation process of the basic power grid topology, so that the power grid topology structure is changed, freezing the weight values and the parameter values of the convolution layer part in the basic model containing the optimal neural network parameters in the step 6, and not freezing the full connection layer;
step 9, aiming at the changed power grid topological structure, generating sample data at the moment by using the method in the step 1, preprocessing the sample data according to the method in the step 2, and then labeling the sample according to the method in the step 3;
step 10, dividing the training set and the test set of the sample data in the step 9 according to the method in the step 4;
step 11, retraining the network parameters of the rest unfrozen parts in the step 8 by using the training set in the step 10 to obtain the corresponding optimal network parameters of the 1D convolution deep learning model when the power grid topological structure is changed;
step 12, testing the model obtained after the training in the step 11 by using the test sample data set in the step 10, if the overfitting phenomenon does not occur, deploying the model into an actual power grid, otherwise, testing the model obtained after the training in the step 11 by using the test sample data set in the step 10 again until the overfitting phenomenon does not occur;
step 13: and deploying the optimal model obtained after training into an actual power grid, inputting the brand new data which is actually monitored, and outputting the positioning result of the voltage sag source.
The invention is also characterized in that:
the specific process of the step 1 is as follows:
taking an IEEE39 node network model as a basic model, modeling by using Matlab/simulink simulation software to generate simulation data, positioning 34 lines where a common voltage sag source can be located, selecting 4 monitoring nodes, analyzing voltage amplitudes of the selected 4 monitoring nodes, positioning the lines where the voltage sag source can be located, sampling the voltage amplitudes of the selected nodes, acquiring voltage amplitude sampling data, and normalizing the data as shown in a formula (1) to obtain preprocessed data:
Figure BDA0003464587250000041
wherein x is*For normalized data output, xmaxIs the maximum value, x, in the input sample dataminIs the minimum value in the input sample data.
The specific process of the step 3 is as follows:
and (3) carrying out manual sample labeling on the data obtained after the processing in the step (2), wherein the input data of each sample is the three-phase voltage amplitude data of the 4 monitoring nodes preprocessed in the step (2), and the output data is a line sequence where a specific voltage sag source in the power distribution network is located, namely 34 lines in the IEEE39 nodes, and is represented by numbers 1-34.
The specific process of the step 5 is as follows:
the deep learning model based on the 1D convolution comprises a convolution layer, a pooling layer and a full-connection layer;
X＝[x1，x2，...，xt，...，xs]Tand (3) transmitting the data to an input layer as model input, wherein X is a voltage time sequence acquired by 4 monitoring nodes after preprocessing in the step (2), s is the length of the time sequence, and the sequence data is mapped into a convolutional layer through one-dimensional convolution operation:
ac j＝fr(X*WC j+b) (2)；
fr(z)＝max(z,0) (3)；
wherein, denotes a one-dimensional convolution operation; a isc jIs represented by a convolution kernel WC jThe generated jth feature map; convolution kernel WC jIs a weight matrix; b is an offset; f. ofr(z) is an activation function.
The specific process of the step 6 is as follows:
training the model constructed in the step 5 by using the training data in the step 4, updating parameters by using a back propagation algorithm, training by using an Adam optimizer, wherein the loss function is a cross soil moisture loss function, the input data is the voltage amplitude sample data of the monitoring node processed in the step 2, and the output data is the line number 1-34 of the short-circuit fault in the step 1.
The method has the beneficial effects that the method is based on the combination of the 1D convolution deep learning model and the migration learning, and the specific line where the voltage sag source is located in the power distribution network is obtained after the three-phase monitoring voltages corresponding to the limited monitoring nodes collected by the power quality monitoring equipment are analyzed. The positioning accuracy of the voltage sag source of the model during topology change is improved, and the self-adaptive capacity and the generalization performance of the model are improved.
Drawings
FIG. 1 is a general flow chart of the present invention for a deep learning based voltage sag source positioning method in view of time varying topology;
FIG. 2 is an iterative update diagram between the generation number and the loss value in the model training process in the deep learning-based voltage sag source positioning method in consideration of the time-varying topology of the present invention;
fig. 3 is an iterative update diagram between the generation number and the accuracy in the model training process of the deep learning-based voltage sag source positioning method in consideration of the time-varying topology 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 relates to a deep learning-based voltage sag source positioning method in time-varying topology, which comprises the following specific steps as shown in fig. 1:
step 1, selecting a power grid topology in a normal operation state as a basic power grid topology. And then, respectively sampling the amplitude values of the three-phase voltages of the nodes by using the electric energy quality monitoring equipment or any other device capable of monitoring the voltage of the power grid node, which is installed in the basic power grid topology. When a single-phase short circuit, a two-phase short circuit and a three-phase short circuit occur in the system, respectively changing the short circuit capacity, the grounding resistance, the duration, the starting and stopping time, the load capacity and the line impedance of the system, and collecting bus voltage data under each operating condition as voltage sample data corresponding to the topological structure of the basic power grid;
the specific process of the step 1 is as follows:
an IEEE39 node network model is used as a basic model, Matlab/simulink simulation software is used for modeling to generate simulation data, and a common voltage sag source can be positioned on 34 lines. The invention selects 4 monitoring nodes (bus3, bus8, bus24 and bus 38). And (4) analyzing the voltage amplitudes of the selected 4 monitoring nodes, and positioning a line where the voltage sag source is possibly located. And sampling the voltage amplitude of the selected node to obtain voltage amplitude sampling data. And the data is normalized as shown in formula (1) to obtain preprocessed data.
Figure BDA0003464587250000061
Wherein x*For normalized data output, xmaxIs the maximum value, x, in the input sample dataminIs the minimum value in the input sample data.
Step 2: calculating the voltage root mean square value of the data acquired in the step 1, and forming a group of voltage sag data samples if the voltage root mean square value is reduced to 90% -10% of a rated value; otherwise, kicking off the sample data acquired in the step 1 to obtain the preprocessed monitoring voltage data.
And step 3: and (3) carrying out sample labeling on the data obtained in the step (2). The input data of each sample is the three-phase voltage amplitude data of the monitoring node preprocessed in the step 2, and the output data is a specific voltage sag source line sequence in the power distribution network.
The specific process of the step 3 is as follows:
and (3) carrying out manual sample labeling on the data obtained after the processing in the steps 1 and 2. The input data of each sample is three-phase voltage amplitude data of 4 monitoring nodes (bus3, bus8, bus24 and bus38) preprocessed in the step 2, and the output data is a line sequence where a specific voltage sag source is located in the power distribution network, namely 34 lines in the IEEE39 nodes, and are represented by numbers 1 to 34.
And 4, step 4: and 4, randomly dividing the sample data in the step 3, wherein 80% of the sample data is used as a training sample data set, and the rest 20% of the sample data is used as a test sample data set.
And 5: and building a 1D convolution deep learning model as a basic model. The 1D convolution deep learning model is a special neural network used for processing sequence data in the step 2 and consists of a convolution layer, a pooling layer and a full-link layer. The convolutional layer can obtain a group of optimal convolutional kernels meeting the minimum loss function through training, and automatic feature extraction is realized by utilizing the convolutional kernels. The pooling layer may extract the most prominent features from the convolutional layer and perform dimension reduction operations in the time dimension. And stacking the convolution layer and the pooling layer to form a deep network structure, and abstracting high-level sequence characteristics layer by layer.
The specific process of the step 5 is as follows:
the method is mainly realized based on a 1D convolution deep learning model and comprises a convolution layer, a pooling layer and a full-connection layer. The convolutional layer can obtain a group of optimal convolutional kernels meeting the minimum loss function through training, and automatic feature extraction is realized by utilizing the convolutional kernels. The pooling layer may extract the most prominent features from the convolutional layer and perform dimension reduction operations in the time dimension. And stacking the convolution layer and the pooling layer to form a deep network structure, and abstracting high-level sequence characteristics layer by layer.
X＝[x1，x2，...，xt，...，xs]TPassed as model input to the input layer. Wherein, X is the voltage acquisition time series of the 4 monitoring nodes (bus3, bus8, bus24 and bus38) preprocessed in the step 2, and s is the length of the time series. The sequence data is mapped into convolutional layers through one-dimensional convolution operation:
ac j＝fr(X*WC j+b) (3)
fr(z)＝max(z,0) (4)
wherein, represents a one-dimensional convolution operation; a isc jIs represented by a convolution kernel WC jThe generated jth feature map; convolution kernel WC jIs a weight matrix; b is an offset; f. ofrAnd (z) is an activation function, which is used for carrying out nonlinear transformation on the data after convolution operation, wherein a ReLu activation function is adopted to accelerate model convergence and enhance sparse representation of the model.
Pooling operations are used to capture the most useful information of convolutional layer sequence features, forming a pooling layer. The pooling operation is typically maximal pooling, and the sequence length can be halved. And when the pooling operation is adopted for the last time, the global maximum pooling is adopted, the most useful global time sequence information is captured, and the sequence length is reduced to 1.
The full connection layer is consistent with the structure of the traditional neural network and consists of a plurality of hidden layers. The fully connected layer further abstractly combines the global timing characteristics.
In the present invention, 2 convolutional layers are used, the number of the first convolutional kernels is 512, the number of the second convolutional kernels is 34, and the number of the neurons in the fully-connected layer is 64. The learning rate is 0.0001 with no extra learning rate attenuation settings.
Step 6: and (5) training the model structure in the step (5) by using the training sample data set in the step (4), updating parameters by using a back propagation algorithm, training by using an Adam optimizer, and obtaining the trained optimal neural network parameters of the basic model, wherein the loss function is a cross soil moisture loss function.
The specific process of the step 6 is as follows:
and (3) training the model in the step (5) by using the training data in the step (4), updating parameters by using a back propagation algorithm, and training by using an Adam optimizer, wherein the loss function is a cross dead-soil loss function. And (3) inputting data which are sample data of the voltage amplitude of the monitoring node processed in the step (2), and outputting data which are line numbers 1-34 of the short-circuit fault in the step (1).
In the whole training process of the model, the super-parameter value needs to be changed to adjust the model, and the optimal model is obtained. The number of layers of the convolutional network, the number of neurons, the type of an activation function and the learning rate can be adjusted.
And 7: and (4) testing the model obtained after the training in the step (6) by using the test sample data set in the step (4), and if an overfitting phenomenon does not occur, taking the model as a final basic model. Otherwise, the model obtained after the training in the step 6 is tested by reusing the test sample data set in the step 4 until the overfitting phenomenon does not occur.
And 8: compared with the basic power grid topology in the step 1, once the power grid topology structure changes due to the fact that one or more branches change in the operation process of the basic power grid topology due to faults or other reasons, the weight values and the parameter values of the convolution layer part in the basic model containing the optimal neural network parameters in the step 6 are frozen, and the full connection layer is not frozen.
The specific process of step 8 is:
and migrating the voltage sag source characteristics under the basic topology to the characteristics after the power grid topology changes by using migration learning. The parameters of model migration mainly comprise the number of neural network layers, the number of neurons and the number of freezing layers.
And step 9: and (3) generating sample data at the moment by using the method in the step 1 for the changed power grid topological structure, preprocessing the sample data according to the method in the step 2, and then labeling the sample according to the method in the step 3.
Step 10: and 4, dividing the training set and the test set of the sample data in the step 9 according to the method in the step 4.
Step 11: and (5) retraining the network parameters of the rest parts which are not frozen in the step 8 by using the training set in the step 10 to obtain the optimal network parameters of the 1D convolution deep learning model corresponding to the changed power grid topological structure.
Step 12: and (3) testing the model obtained after the training in the step (11) by using the test sample data set in the step (10), and deploying the model to an actual power grid if the overfitting phenomenon does not occur. Otherwise, the model obtained after the training in the step 11 is tested by reusing the test sample data set in the step 10 until the overfitting phenomenon does not occur.
Step 13: and deploying the optimal model obtained after training into an actual power grid, inputting the brand new data which is actually monitored, and outputting the positioning result of the voltage sag source.
The specific process of step 13 is:
and (4) taking the optimal network parameters of the 1D convolution deep learning model obtained in the step (11) as a basic model to be deployed in an actual power grid. And once the topological structure of the power grid is changed, offline fine adjustment is carried out on the basic network by using a small amount of new topological historical data. If the test accuracy of the model after fine tuning meets the requirement, real-time voltage measurement data after topology change is acquired on line, 4 monitoring nodes (bus3, bus8, bus24 and bus38) are input to the model after fine tuning, and the number of the line where the specific voltage sag source is located in the power grid is output to be 1-34.
The following explanation is made with respect to the drawings:
FIG. 1 shows the general process steps of the present invention, including two phases of off-line training and on-line positioning.
Fig. 2 is an iteration update diagram between the generation number and the loss value in the model training process when 0 branch, 1 branch, 3 branches, and 5 branches are respectively disconnected in the IEEE39 node system, and the model loss value gradually decreases with the increase of the iteration number.
Fig. 3 is an iteration update diagram between the generation number and the accuracy in the model training process when 0 branch, 1 branch, 3 branches, and 5 branches are respectively disconnected in the IEEE39 node system, and the model accuracy is improved to 99.3% as the iteration number increases.

Claims (5)

1. The voltage sag source positioning method based on deep learning in the time-varying topology is characterized by comprising the following steps: the method specifically comprises the following steps:
step 1, collecting bus voltage data under different operating conditions, and taking the sample data as sample data corresponding to a basic power grid topological structure;
step 2, calculating the voltage root mean square value of the sample data acquired in the step 1, and if the voltage root mean square value is reduced to 90% -10% of a rated value, forming a group of voltage sag data samples by the sample data corresponding to the voltage root mean square value; otherwise, kicking off the sample data from the sample data acquired in the step 1 to obtain the preprocessed monitoring voltage data;
step 3, carrying out sample marking on the data obtained in the step 2, wherein the input data of each sample is the three-phase voltage amplitude data of the monitoring node preprocessed in the step 2, and the output data is a line sequence where a specific voltage sag source is located in the power distribution network;
step 4, randomly dividing the sample data marked in the step 3, wherein 80% of the sample data is used as a training sample data set, and the rest 20% is used as a test sample data set;
step 5, building a 1D convolution deep learning model as a basic model;
step 6, training the model set up in the step 5 by using the training sample data set in the step 4, updating parameters by using a back propagation algorithm, training by using an Adam optimizer, and obtaining the optimal neural network parameters of the trained basic model, wherein the loss function is a cross soil moisture loss function;
step 7, testing the model obtained after the training in the step 6 by using the test sample data set in the step 4, and if an overfitting phenomenon does not occur, taking the model as a final basic model; otherwise, testing the model obtained after the training in the step 6 by using the test sample data set in the step 4 again until no overfitting phenomenon occurs;
step 8, comparing with the basic power grid topology in the step 1, once one or more branches are changed due to faults in the operation process of the basic power grid topology, so that the power grid topology structure is changed, freezing the weight values and the parameter values of the convolution layer part in the basic model containing the optimal neural network parameters in the step 6, and not freezing the full connection layer;
step 9, aiming at the changed power grid topological structure, generating sample data at the moment by using the method in the step 1, preprocessing the sample data according to the method in the step 2, and then labeling the sample according to the method in the step 3;
step 10, dividing the training set and the test set of the sample data in the step 9 according to the method in the step 4;
step 11, retraining the network parameters of the rest unfrozen parts in the step 8 by using the training set in the step 10 to obtain the corresponding optimal network parameters of the 1D convolution deep learning model when the power grid topological structure is changed;
step 12, testing the model obtained after the training in the step 11 by using the test sample data set in the step 10, if the overfitting phenomenon does not occur, deploying the model into an actual power grid, otherwise, testing the model obtained after the training in the step 11 by using the test sample data set in the step 10 again until the overfitting phenomenon does not occur;
step 13: and deploying the optimal model obtained after training into an actual power grid, inputting the brand new data which is actually monitored, and outputting the positioning result of the voltage sag source.
2. The method for positioning the voltage sag source based on deep learning in the time-varying topology consideration of the step 1 is characterized in that: the specific process of the step 1 is as follows:
taking an IEEE39 node network model as a basic model, modeling by using Matlab/simulink simulation software to generate simulation data, positioning a common voltage sag source on 34 lines, selecting 4 monitoring nodes, analyzing voltage amplitudes of the selected 4 monitoring nodes, positioning the lines where the voltage sag source may be located, sampling the voltage amplitudes of the selected nodes, acquiring voltage amplitude sampling data, and normalizing the data as shown in formula (1) to obtain preprocessed data:
Figure FDA0003464587240000031
wherein x is*For normalized data output, xmaxIs the maximum value, x, in the input sample dataminIs the minimum value in the input sample data.
3. The method for positioning the voltage sag source based on deep learning in consideration of the time-varying topology in step 2 is characterized in that: the specific process of the step 3 is as follows:
and (3) carrying out manual sample labeling on the data obtained after the processing in the step (2), wherein the input data of each sample is the three-phase voltage amplitude data of the 4 monitoring nodes preprocessed in the step (2), and the output data is a line sequence where a specific voltage sag source in the power distribution network is located, namely 34 lines in the IEEE39 nodes and is represented by numbers 1-34.
4. The method for positioning the voltage sag source based on deep learning in consideration of the time-varying topology in step 3 is characterized in that: the specific process of the step 5 is as follows:
the deep learning model based on the 1D convolution comprises a convolution layer, a pooling layer and a full-connection layer;
X＝[x1，x2，...，xt，...，xs]Tand (3) transmitting the data to an input layer as model input, wherein X is a voltage time sequence acquired by 4 monitoring nodes after preprocessing in the step (2), s is the length of the time sequence, and the sequence data is mapped into a convolutional layer through one-dimensional convolution operation:
ac j＝fr(X*WC j+b) (2)；
fr(z)＝max(z,0) (3)；
wherein, represents a one-dimensional convolution operation; a isc jIs represented by a convolution kernel WC jThe generated jth feature map; convolution kernel WC jIs a weight matrix; b is an offset; f. ofr(z) is an activation function.
5. The method for positioning the voltage sag source based on deep learning in consideration of the time-varying topology in step 4 is characterized in that: the specific process of the step 6 is as follows:
training the model constructed in the step 5 by using the training data in the step 4, updating parameters by using a back propagation algorithm, training by using an Adam optimizer, wherein the loss function is a cross soil moisture loss function, the input data is the voltage amplitude sample data of the monitoring node processed in the step 2, and the output data is the line number 1-34 of the short-circuit fault in the step 1.
CN202210025828.7A 2022-01-11 2022-01-11 Voltage sag source positioning method based on deep learning in consideration of time-varying topology Pending CN114545147A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210025828.7A CN114545147A (en) 2022-01-11 2022-01-11 Voltage sag source positioning method based on deep learning in consideration of time-varying topology

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210025828.7A CN114545147A (en) 2022-01-11 2022-01-11 Voltage sag source positioning method based on deep learning in consideration of time-varying topology

Publications (1)

Publication Number Publication Date
CN114545147A true CN114545147A (en) 2022-05-27

Family

ID=81669068

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210025828.7A Pending CN114545147A (en) 2022-01-11 2022-01-11 Voltage sag source positioning method based on deep learning in consideration of time-varying topology

Country Status (1)

Country Link
CN (1) CN114545147A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115713158A (en) * 2022-11-23 2023-02-24 贵州电网有限责任公司信息中心 Power distribution network stability prediction method, device, equipment and storage medium

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115713158A (en) * 2022-11-23 2023-02-24 贵州电网有限责任公司信息中心 Power distribution network stability prediction method, device, equipment and storage medium

Similar Documents

Publication Publication Date Title
CN109873610B (en) Photovoltaic array fault diagnosis method based on IV characteristic and depth residual error network
CN109459669B (en) 10kV power distribution network single-phase earth fault section positioning method
CN109672221B (en) Direct-drive wind power plant dynamic equivalence method for subsynchronous oscillation analysis
CN110570122B (en) Offshore wind power plant reliability assessment method considering wind speed seasonal characteristics and current collection system element faults
CN111475909B (en) Wind turbine generator output correlation mapping modeling method based on long-term and short-term memory network
CN103728535A (en) Extra-high-voltage direct-current transmission line fault location method based on wavelet transformation transient state energy spectrum
CN112787591B (en) Photovoltaic array fault diagnosis method based on fine-tuning dense connection convolutional neural network
CN111308260B (en) Electric energy quality monitoring and electric appliance fault analysis system based on wavelet neural network and working method thereof
CN110059737B (en) Distribution transformer connection relation identification method based on integrated deep neural network
CN114545147A (en) Voltage sag source positioning method based on deep learning in consideration of time-varying topology
CN111999591B (en) Method for identifying abnormal state of primary equipment of power distribution network
CN115392133A (en) Wind power plant optimal clustering equivalence method adopting Gaussian mixture model
CN113804997B (en) Voltage sag source positioning method based on bidirectional WaveNet deep learning
CN115146538A (en) Power system state estimation method based on message passing graph neural network
CN113514743B (en) Method for constructing GIS partial discharge pattern recognition system based on multidimensional features
Hassani et al. Design of a cost-effective deep convolutional neural network–based scheme for diagnosing faults in smart grids
CN116973683A (en) Method and system for positioning single-phase ground fault section of electric collecting line of wind power plant
CN115795360A (en) Cable fault detection method based on artificial neural network
CN116317937A (en) Distributed photovoltaic power station operation fault diagnosis method
CN115856530A (en) Edge side online arc fault detection method and system based on multidimensional entropy characteristics
CN115659133A (en) Wind power converter fault diagnosis method
CN111177973B (en) Photovoltaic array online modeling method based on reinforcement learning
CN115293090A (en) Multi-harmonic source responsibility quantification method based on reconstruction data processing
CN114626957A (en) Voltage sag state evaluation method based on gate control cycle unit deep learning model
CN114035120A (en) Three-level inverter open-circuit fault diagnosis method and system based on improved CNN

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