CN115877068A - Voltage sag propagation track identification method of regional power grid based on deep learning - Google Patents
Voltage sag propagation track identification method of regional power grid based on deep learning Download PDFInfo
- Publication number
- CN115877068A CN115877068A CN202210762648.7A CN202210762648A CN115877068A CN 115877068 A CN115877068 A CN 115877068A CN 202210762648 A CN202210762648 A CN 202210762648A CN 115877068 A CN115877068 A CN 115877068A
- Authority
- CN
- China
- Prior art keywords
- sag
- power grid
- propagation
- convolution
- data
- 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
- Y04S10/52—Outage or fault management, e.g. fault detection or location
Landscapes
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention discloses a deep learning-based voltage sag propagation track identification method for a regional power grid, which comprises the following steps of: step 1: constructing a fusion recognition model which consists of SAE dimension reduction and AttUnet classification recognition; and 2, step: constructing a large-scale monitoring point parallel real-time mode recognition platform facing a regional power grid based on Flink, and realizing integration of sag data acquisition, processing and storage; and 3, step 3: and constructing a propagation track amplitude estimation model based on GAT, and converting track data into trainable graph structure data through a power grid adjacency matrix. The method can estimate the amplitude change of the node which is not provided with the monitoring device in the sag propagation process, has higher generalization capability and precision, and plays an important role in reducing financial, human and material resource investment on the deployment of the monitoring device to a certain extent for a power grid system.
Description
Technical Field
The invention relates to a power grid, in particular to a voltage sag propagation track identification method facing to a regional power grid based on deep learning.
Background
With the further development of the interconnection of the power grid, the disturbance propagation phenomenon of the power system is more prominent, various mechanical wave expression forms are presented, if the propagation process of the disturbance in the power grid cannot be correctly analyzed, so that a proper control strategy is proposed, cascading failures are very likely to be caused, and even large-area power failure can be caused in the worst case, because the power transmission process usually experiences a plurality of transformers and transmission lines, the voltage sag characteristic of a client equipment terminal is possibly greatly different from the voltage sag characteristic in the failure, the voltage sag characteristic on the power transmission network can be monitored and evaluated, but in the industrial facility of a client, the voltage sag characteristic of a network terminal is usually required to be calculated according to the propagation rule of the voltage sag characteristic, so that the important significance is achieved in researching the propagation rule of the voltage sag, and the propagation of the voltage sag on the power grid is divided into transmission line propagation and transformer propagation, both conform to the superposition theorem, wherein the propagation on the transmission line is only related to the impedance of the line, the propagation on the transformer is related to the voltage sag type, the connection and grounding mode of the transformer and the winding, in the power system, each generator and transformation bus (node) are connected into a small-world network through the transmission line, the topological structure of the power grid in operation has a significant influence on the dynamic characteristic of the system and the sag propagation, the power quality pollution has a propagation characteristic in the power grid, resulting in a coupling effect between the nodes, especially for adjacent nodes, the power quality pollution between the nodes may show a strong coupling relation due to the similar positions of the adjacent nodes relative to the pollution source, the amplitude and duration are the main characteristics of the voltage sag, but the starting point and the duration of the voltage sag are not changed in the propagation process, therefore, the change situation of the voltage sag amplitude in the regional power grid propagation process becomes a difficult point for research.
For a complex voltage sag event, the existing monitoring system adopts an independent monitoring mode of monitoring points, collected data are discrete and isolated, real-time disturbance analysis of multiple monitoring points cannot be realized facing a regional power grid, complete propagation and comprehensive information of complex disturbance cannot be fully expressed, the same sag propagation track cannot be extracted from the complete sag propagation process facing the regional power grid, when the sag event occurs in the power grid, how to face the regional power grid, based on the dynamic correlation characteristics of all nodes under the same time section, the extraction of the same sag propagation track in the power grid becomes a key for researching sag propagation characteristics, the traditional data characterization learning and identification algorithm is difficult to process a large sample sag data set or when the sample volume and the feature vector dimension are huge, higher algorithm complexity can affect the identification effect, the existing monitoring system is limited to real-time monitoring of a single monitoring point, real-time correlation analysis and calculation of multiple monitoring points in a complex regional power grid environment cannot be faced, and efficient and accurate sag identification algorithm and real-time correlation analysis facing the regional are the premise of extracting the same sag propagation track.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a voltage sag propagation track identification method facing a regional power grid based on deep learning.
A voltage sag propagation track identification method facing a regional power grid based on deep learning comprises the following steps:
step 1: constructing a fusion recognition model, comprising SAE dimension reduction and AttUnet classification recognition, inputting a temporary drop three-phase signal, wherein an SAE encoder is two convolution layers, a hidden layer is a full-connection layer and is optimized by a sparse constraint unit, a decoder is two anti-convolution layers and one convolution layer, each layer is normalized by Batch Normalization (BN) and is activated by ReLU, the low-dimensional characteristics of the hidden layer enter the AttUnet recognition, the AttUnet encoder is 2 convolution blocks and 2 down-sampling layers, the convolution blocks are composed of 2 convolution layers, the decoder is 3 convolution layers and 2 up-sampling layers, AG is composed of 1-dimensional convolution, reLU and Sigmoid to synthesize encoding and decoding output characteristics and substitutes Softmax for classification, a loss function is a cross entropy loss function, and the temporary drop event classification is realized by training multiple times to minimize loss until Nash balance is reached;
step 2: constructing a large-scale monitoring point parallel real-time mode recognition platform facing a regional power grid based on Flink, realizing integration of sag data acquisition, processing and storage, acquiring characteristic data of each monitoring point when sag occurs in real time by using Flume, forwarding the characteristic data to a Flink cluster through a Kafka cluster for processing, extracting sag tracks based on a sliding window, uniformly managing the cluster through a Zookeeper, storing the extracted track data through a database for researching propagation characteristics, wherein the sliding window algorithm is as follows: the sag data of the whole network monitoring points enter a sliding window, all monitoring points of a regional power grid corresponding to a basic window are processed one by one, sag types are identified firstly, then aggregation judgment is carried out, if the same sag event is captured by at least Z monitoring points in T seconds, the sag data serve as a propagation track of the event in the power grid, a plurality of monitoring point information tuple arrays corresponding to the event are stored in a database, and each tuple corresponds to single monitoring point information and consists of a number, an amplitude value, a sag phase and a bus access power supply condition;
and step 3: a propagation track amplitude estimation model based on GAT is constructed, track data are converted into trainable graph structure data through a power grid adjacency matrix, central nodes are updated through 2 layers of 8-head and 4-head GAT networks and adjacent edge node features are aggregated through a splicing and averaging mode, then the new features are substituted into 2 layers of full connection layers, output dimensions are 5 and 1 respectively, estimated amplitudes are output, the model is activated through eLU, tanh and BN, a loss function is optimized through a smoothL1 piecewise function, errors between the estimated amplitudes and real amplitudes are minimized through multiple training, and amplitude estimation of nodes without monitoring devices in sag propagation is finally achieved.
Optionally, the method further includes: and 4, step 4: verifying the estimation capability of the model, namely, referring the representative nodes of the power grid with the monitoring devices removed as key nodes, wherein the key nodes are standardized to a connection relationship, namely, aggregation degree, electrical distances among the nodes, and electrical distances among the nodes and a power supply, namely, electrical distances and support levels, quantitative indexes are sequences formed by average voltage sag amplitudes of n key nodes in a regional power grid in a monitoring period, namely, a voltage sag mode, simulating sag propagation experiments for 15 times, generating 100 faults in each experiment, and obtaining the influence rule of the power grid structure on sag propagation, namely, the propagation characteristics by comparing the sag modes of the key nodes under different power grid structures.
Further, the step 1 comprises: step 11: the SAE input layer is original data A, the encoder is two convolution layers, the convolution kernel sizes are 3 x 3, the step lengths are 1 and 2 in sequence, the hidden layer is a full-connection layer and is optimized by a sparse constraint unit, the decoder is two anti-convolution layers and one convolution layer, the convolution kernel size is 2 x 2, the step length is 1, the decoder output is B, the model minimizes the error between A and B, and the pre-trained SAE dimension reduction model takes the hidden layer low-dimensional feature C as the result output; step 12: the characteristic C enters AttUnet for identification, an AttUnet encoder comprises 2 convolution blocks and 2 down-sampling layers, each convolution block consists of 2 convolution layers with convolution kernel size of 3 x 3, the down-sampling layer is a maximum pooling layer with convolution kernel size of 2 x 2, the characteristic D is extracted downwards, a decoder comprises 3 convolution layers and 2 up-sampling layers, the convolution kernel sizes of the convolution layers are 3 x 3, the convolution kernel size of the up-sampling layer is 2 x 2, the characteristic E of upward expansion is E, the AttUnet network has 2 times of cross-layer connection, the characteristics of D and E are integrated by AG before connection, the D and E enter Softmax to obtain various category distribution probabilities, and further temporary drop identification and classification are carried out.
Wherein the step 2 comprises: the method comprises the steps that sag data collected by Flume in real time are uploaded to a Flink cluster through Kafka to be subjected to track extraction, monitoring point data obtained by a time sliding window are dispersed in a basic window in a memory, amplitude calculation, sag identification and data integration are carried out concurrently, then the data are subjected to aggregation analysis, if the data are the same sag event, T is 1.5s, and Z is 7, the condition shows that only if the sag event is captured in a time section of 1.5s and at least 7 monitoring points exist, the monitoring point information can be stored as a propagation track of the sag event.
The invention provides a sag identification method fusing multilayer SAE and AttUnet, wherein SAE is used for training the mapping relation from the sequence of an encoder and a decoder to the sequence, implicit layer vector space characteristics are used as low-dimensional high-precision characteristic expression to enter AttUnet for further extracting characteristics and identifying, the AttUnet introduces global context information, the encoder is used for extracting bottom layer characteristics in a downward depth mode, the decoder is used for expanding the bottom layer characteristics upwards to form high layer characteristics, and the extracted characteristics have high discrimination and expressive force through cross-layer AG combination, so that the identification precision is improved; then constructing a parallel real-time mode recognition platform facing a large-scale monitoring point of a regional power grid based on Flink, wherein the platform can process unbounded data traffic in real time, in parallel and continuously, and providing a propagation track extraction algorithm based on a sliding time window to obtain the propagation track characteristics of sag events in real time for propagation characteristic research; a propagation track amplitude estimation model based on GAT is provided, monitoring point information in track data is converted into trainable graph structure data through one-to-one correspondence of adjacent matrixes, influence of each monitoring point is different due to different sag levels, the GAT is combined with an attention module, and correlation among nodes is concerned while the trace graph structure data is learned, so that amplitude estimation of key nodes in sag propagation is achieved; the aggregation degree, the electrical distance and the support level are different and reflect different power grid structures, key nodes are selected based on the aggregation degree, the electrical distance and the support level, the amplitude of each node is estimated by a model, and the influence rule of the power grid topology on sag propagation, namely the propagation characteristic, is obtained by comparing the amplitude change conditions of the key nodes under different structures.
The invention adopts SAE, attUnet and GAT with powerful autonomous feature extraction capability as main model architecture, and Flink as a track extraction main computing platform to realize multi-point real-time mode identification and sag propagation rule discovery.
The invention adopts GAT training trajectory diagram structural data, fuses the time-space correlation of the power grid space structure, can estimate the amplitude change of the node which is not provided with the monitoring device in the sag propagation process, has higher generalization capability and precision, and has important effect on reducing financial, human and material resource investment on the deployment of the monitoring device for a certain extent in a power grid system.
Drawings
Fig. 1 is an overall technical configuration diagram.
Fig. 2 is a sag recognition model fusing SAE and AttUnet.
FIG. 3 is a parallel real-time pattern recognition platform facing a large-scale monitoring point of a regional power grid based on Flink.
FIG. 4 is a voltage sag propagation trajectory extraction model based on a time sliding window.
Fig. 5 is a GAT-based propagation trajectory amplitude estimation model.
Detailed Description
The above and other objects, features and advantages of the present invention will become more apparent from the following detailed description of the preferred embodiments of the present invention when taken in conjunction with the accompanying drawings. Like reference numerals refer to like parts throughout the drawings. The drawings are not intended to be to scale, emphasis instead being placed upon illustrating the principles of the invention.
The invention provides a deep learning-based regional power grid-oriented voltage sag propagation track and characteristic research method, wherein a real-time track of voltage sag propagation is a data basis of propagation characteristic research, accurate voltage sag identification is a key for extracting the voltage sag propagation track, and the method comprises the following operation steps: the method comprises the steps that firstly, a regional power grid is required to be oriented to collect mass sag information to carry out sag event classification training, high-dimensional sag characteristics contain rich power system operation state information, and the sag identification method has obvious influence on high-precision identification; then constructing a Flink-based large-scale monitoring point parallel real-time mode recognition platform facing the regional power grid, wherein the platform can process unbounded data traffic in real time, in parallel and continuously, and providing a propagation track extraction algorithm based on a sliding time window, so that propagation track characteristics of sag events are obtained in real time and are used for propagation characteristic research; providing a propagation track amplitude estimation model based on a Graph Attention neural Network (GAT), wherein monitoring point information in track data is converted into trainable Graph structure data through one-to-one correspondence of adjacent matrixes, and the influence of each monitoring point is different due to different transient levels; the aggregation degree, the electrical distance and the support level are different and reflect different power grid structures, key nodes are selected based on the aggregation degree, the electrical distance and the support level, the amplitude of each node is estimated by a model, and the influence rule of the power grid topology on sag propagation, namely the propagation characteristic, is obtained by comparing the amplitude change conditions of the key nodes under different structures.
A sag identification method fusing a multilayer Sparse Auto-Encoder (SAE) and an attentionUnnet (AttUnnet) is characterized in that a regional power grid is oriented to collecting mass sag information to conduct sag event classification training, high-dimensional sag characteristics contain rich power system operation state information, and the sag characteristics have obvious influence on high-precision identification.
As shown in fig. 1, the existing research mainly analyzes the change rule of a single event by a transformer and a line based on an isolated time section and a typical working condition through mechanism analysis, sag propagation can be regarded as a result of combined action of disturbance and power grid topology, when a sag occurs in a regional power grid, a monitoring point can instantly capture the sag event, an obvious amplitude change occurs, only the sag type is accurately identified, and then the sag event propagation trajectory is completely tracked in real time, and based on dynamic trajectory data, the influence rule of a power grid structure on the sag propagation is obtained, an efficient sag recognition model is a key for correctly extracting the propagation trajectory, and trajectory data extracted by a real-time mode recognition platform is a data base for researching the propagation characteristic, and the propagation characteristic is researched through the invention.
As shown in fig. 2, the recognition model of this embodiment is classified into SAE feature dimension reduction and AttUnet classification, 900 × 3 high dimension temporary reduction is compressed by an SAE encoder to a decoder to be reconstructed to realize training from a sequence to a sequence, and as a result, the low dimension features of the hidden layer 256 × 1 enter the AttUnet for classification, so as to reduce the complexity of classification training.
As shown in fig. 3, the real-time pattern recognition platform of this embodiment realizes integration of data acquisition, processing and storage, where the flute real-time acquisition node temporarily drops data and forwards the data to the Flink cluster via Kafka for processing, and performs sliding window trajectory extraction, the cluster is uniformly managed by Zookeeper, and the extracted trajectory data is stored in the database for propagation characteristic research.
As shown in fig. 4, in the time sliding window trajectory extraction algorithm based on Flink according to this embodiment, when a voltage sag occurs in a power grid, sag data of monitoring points may enter a sliding window for logic analysis, the sliding window is composed of basic windows, the basic windows correspond to monitoring nodes in the power grid system, the sliding window slides forward every T seconds along a data stream direction, the basic windows in the window correspond to newly arrived monitoring node data and are analyzed by a service logic processing component, if a sag event is captured by at least Z monitoring points in a time segment T, a propagation trajectory generated by propagation of the sag event in the power grid is obtained, and finally information data of a plurality of monitoring points affected by the sag event is integrated as a propagation trajectory and recorded in a database.
As shown in fig. 5, in the GAT-based propagation trajectory amplitude estimation model of this embodiment, trajectory data and a power grid structure are mapped into trainable graph structure data, a coupling relationship between adjacent edge node features and a central node amplitude is obtained through two steps of feature extraction and amplitude estimation, and finally, amplitude estimation of a central node to which a monitoring device is not installed is achieved.
The invention provides a voltage sag propagation track and characteristic research method facing a regional power grid based on deep learning, which belongs to the technical field of large electric power data, and the method comprises the steps of firstly, collecting mass sag information facing the regional power grid to perform sag event classification training, wherein high-dimensional sag characteristics contain rich electric power system operation state information and have obvious influence on high-precision identification; then constructing a parallel real-time mode recognition platform facing a large-scale monitoring point of a regional power grid based on Flink, wherein the platform can process unbounded data traffic in real time, in parallel and continuously, and providing a propagation track extraction algorithm based on a sliding time window to obtain the propagation track characteristics of sag events in real time for propagation characteristic research; a propagation track amplitude estimation model based on GAT is provided, monitoring point information in track data is converted into trainable graph structure data through one-to-one correspondence of adjacent matrixes, influence of each monitoring point is different due to different sag levels, the GAT is combined with an attention module, and correlation among nodes is concerned while the trace graph structure data is learned, so that amplitude estimation of key nodes in sag propagation is achieved; the aggregation degree, the electrical distance and the support level are different and reflect different power grid structures, key nodes are selected based on the aggregation degree, the electrical distance and the support level, the amplitude of each node is estimated by a model, and the influence rule of the power grid topology on sag propagation, namely the propagation characteristic, is obtained by comparing the amplitude change conditions of the key nodes under different structures.
In the previous description, numerous specific details were set forth in order to provide a thorough understanding of the present invention. The foregoing description is only a preferred embodiment of the invention, which can be embodied in many different forms than described herein, and therefore the invention is not limited to the specific embodiments disclosed above. And that those skilled in the art may, using the methods and techniques disclosed above, make numerous possible variations and modifications to the disclosed embodiments, or modify equivalents thereof, without departing from the scope of the claimed embodiments. Any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present invention are within the scope of the technical solution of the present invention.
Claims (5)
1. A voltage sag propagation track identification method facing a regional power grid based on deep learning is characterized by comprising the following steps:
step 1: constructing a fusion recognition model, which is composed of SAE dimension reduction and AttUnet classification recognition, inputting a sag three-phase signal, an SAE coder is two convolution layers, a hidden layer is a full-connection layer and is optimized by a sparse constraint unit, a decoder is two anti-convolution layers and one convolution layer, each layer is normalized by a Batch Normalization (BN) and is activated by a ReLU, the low-dimensional characteristics of the hidden layer enter the AttUnet recognition, the AttUnet coder is 2 convolution blocks and 2 down-sampling layers, the convolution blocks are composed of 2 convolution layers, the decoder is 3 convolution layers and 2 up-sampling layers, AG is composed of 1-dimensional convolution, reLU and Sigmoid to comprehensively encode and decode the output characteristics and substitute the output characteristics into Softmax for classification, a loss function is a cross entropy loss function, and the sag event classification is realized by training loss for multiple times until Nash balance is achieved;
and 2, step: constructing a large-scale monitoring point parallel real-time mode recognition platform facing a regional power grid based on Flink, realizing integration of sag data acquisition, processing and storage, acquiring characteristic data of each monitoring point when sag occurs in real time by using Flume, forwarding the characteristic data to a Flink cluster through a Kafka cluster for processing, extracting sag tracks based on a sliding window, uniformly managing the clusters through a Zookeeper, storing the extracted track data through a database for researching propagation characteristics, wherein the sliding window algorithm is as follows: the sag data of the whole network can enter a sliding window, all monitoring points of a regional power grid corresponding to a basic window are processed one by one, sag types are firstly identified, then aggregation judgment is carried out, if the same sag event is captured by at least Z monitoring points within T seconds, the sag data is taken as a propagation track of the event on the power grid, a plurality of monitoring point information tuple arrays corresponding to the event are stored in a database, and a single tuple corresponds to single monitoring point information and consists of a number, an amplitude value, a sag phase and a bus access power supply condition;
and step 3: a propagation track amplitude estimation model based on GAT is constructed, track data are converted into trainable graph structure data through a power grid adjacency matrix, central nodes are updated through 2 layers of 8-head and 4-head GAT networks and adjacent edge node features are aggregated through a splicing and averaging mode, then the new features are substituted into 2 layers of full connection layers, output dimensions are 5 and 1 respectively, estimated amplitudes are output, the model is activated through eLU, tanh and BN, a loss function is optimized through smoothL1 piecewise functions, errors between the estimated amplitudes and real amplitudes are minimized through multiple training, and finally amplitude estimation of nodes without monitoring devices in sag propagation is achieved.
2. The identification method according to claim 1, further comprising:
and 4, step 4: verifying the estimation capability of the model, namely, referring the representative nodes of the power grid with the monitoring devices removed as key nodes, wherein the key nodes are standardized to a connection relationship, namely, aggregation degree, electrical distances among the nodes, and electrical distances among the nodes and a power supply, namely, electrical distances and support levels, quantitative indexes are sequences formed by average voltage sag amplitudes of n key nodes in a regional power grid in a monitoring period, namely, a voltage sag mode, simulating sag propagation experiments for 15 times, generating 100 faults in each experiment, and obtaining the influence rule of the power grid structure on sag propagation, namely, the propagation characteristics by comparing the sag modes of the key nodes under different power grid structures.
3. The identification method according to claim 1, wherein the step 1 comprises:
step 11: the SAE input layer is original data A, the encoder is two layers of convolution layers, the sizes of convolution kernels are 3 x 3, the step lengths are 1 and 2 in sequence, the hidden layer is a full-connection layer and is optimized by a sparse constraint unit, the decoder is two layers of anti-convolution layers and one layer of convolution layer, the size of the convolution kernel is 2 x 2, the step length is 1, the output of the decoder is B, the error of the A and B is minimized by the model, and the pre-trained SAE dimension reduction model takes the low-dimensional characteristic C of the hidden layer as the result to be output;
step 12: the characteristic C enters AttUnet for identification, an AttUnet encoder comprises 2 convolution blocks and 2 down-sampling layers, each convolution block consists of 2 convolution layers with convolution kernel size of 3 x 3, the down-sampling layer is a maximum pooling layer with convolution kernel size of 2 x 2, the characteristic D is extracted downwards, a decoder comprises 3 convolution layers and 2 up-sampling layers, the convolution kernel sizes of the convolution layers are 3 x 3, the convolution kernel size of the up-sampling layer is 2 x 2, the characteristic E of upward expansion is E, the AttUnet network has 2 times of cross-layer connection, the characteristics of D and E are integrated by AG before connection, the D and E enter Softmax to obtain various category distribution probabilities, and further temporary drop identification and classification are carried out.
4. The identification method according to claim 1, wherein the step 2 comprises:
the method comprises the steps that pause data collected by Flume in real time are uploaded to a Fl ink cluster through Kafka to be subjected to track extraction, monitoring point data obtained through a time sliding window are dispersed in a basic window in a memory, amplitude calculation, pause identification and data integration are carried out, then the data are subjected to aggregation analysis, if the data are the same pause event, T is 1.5s, and Z is 7, it is shown that only if the pause event is captured by at least 7 monitoring points within 1.5s of a time section, the monitoring point information can be stored as a propagation track of the pause event.
5. The identification method according to claim 1, wherein the number of heads of the GAT is: the single-head GAT means that only one GAT is used for calculating the attention coefficient of the adjacent node to the central node and performing feature aggregation to update the characteristics of the central node, and the multiple heads can integrate different preferences of different GAT networks to enable the characteristics of the central node to be richer in expressiveness.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210762648.7A CN115877068A (en) | 2022-06-29 | 2022-06-29 | Voltage sag propagation track identification method of regional power grid based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210762648.7A CN115877068A (en) | 2022-06-29 | 2022-06-29 | Voltage sag propagation track identification method of regional power grid based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115877068A true CN115877068A (en) | 2023-03-31 |
Family
ID=85769471
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210762648.7A Pending CN115877068A (en) | 2022-06-29 | 2022-06-29 | Voltage sag propagation track identification method of regional power grid based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115877068A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117439068A (en) * | 2023-10-26 | 2024-01-23 | 国网宁夏电力有限公司中卫供电公司 | Voltage sag estimation method, medium and system in large-scale power grid |
| CN117475241A (en) * | 2023-12-27 | 2024-01-30 | 山西省水利建筑工程局集团有限公司 | Geological mutation detection system and method for tunnel excavation of cantilever type heading machine |
-
2022
- 2022-06-29 CN CN202210762648.7A patent/CN115877068A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117439068A (en) * | 2023-10-26 | 2024-01-23 | 国网宁夏电力有限公司中卫供电公司 | Voltage sag estimation method, medium and system in large-scale power grid |
| CN117439068B (en) * | 2023-10-26 | 2024-05-14 | 国网宁夏电力有限公司中卫供电公司 | Voltage sag estimation method, medium and system in large-scale power grid |
| CN117475241A (en) * | 2023-12-27 | 2024-01-30 | 山西省水利建筑工程局集团有限公司 | Geological mutation detection system and method for tunnel excavation of cantilever type heading machine |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110995475B (en) | Power communication network fault detection method based on transfer learning | |
| CN115877068A (en) | Voltage sag propagation track identification method of regional power grid based on deep learning | |
| CN110943857A (en) | Power communication network fault analysis and positioning method based on convolutional neural network | |
| CN112684379A (en) | Transformer fault diagnosis system and method based on digital twinning | |
| CN108665112A (en) | Photovoltaic fault detection method based on Modified particle swarm optimization Elman networks | |
| CN109635928A (en) | A kind of voltage sag reason recognition methods based on deep learning Model Fusion | |
| CN116245033B (en) | Artificial intelligent driven power system analysis method and intelligent software platform | |
| CN112818035A (en) | Network fault prediction method, terminal equipment and storage medium | |
| CN115757813A (en) | Equipment residual life prediction method based on fault time sequence knowledge graph | |
| CN112925824A (en) | Photovoltaic power prediction method and system for extreme weather type | |
| CN118013675B (en) | Hierarchical compression-based power distribution network panoramic topology model construction method and system | |
| CN110689272A (en) | Sensitive equipment tolerance characteristic real-time online evaluation method based on stream computing | |
| Zhang et al. | STPGTN–A Multi-Branch Parameters Identification Method Considering Spatial Constraints and Transient Measurement Data. | |
| CN118133542A (en) | Electric power safety control system based on digital twin | |
| CN113780398A (en) | Wireless network link quality prediction method and system | |
| CN117972419A (en) | Anomaly-transporter photovoltaic power generation abnormality detection method | |
| CN116340830B (en) | Highway electromechanical system fault classification method based on deep memory model | |
| CN117131452A (en) | Abnormality detection method and system based on normalized flow and Bayesian network | |
| CN116167465A (en) | Solar irradiance prediction method based on multivariate time series ensemble learning | |
| CN117154680A (en) | Wind power prediction method based on non-stationary transducer model | |
| CN116644361A (en) | Overhead transmission line fault diagnosis method and computer readable medium | |
| CN116244600A (en) | Method, system and equipment for constructing GIS intermittent discharge mode identification model | |
| CN110348489A (en) | A kind of partial discharge of transformer mode identification method based on autoencoder network | |
| CN115115139A (en) | Real-time prediction method for wind power | |
| CN118413456B (en) | Network abnormity early warning method and early warning system based on AAE |
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 | ||
| CB02 | Change of applicant information |
Address after: 030001 No.6 Qingnian Road, Yingze District, Taiyuan City, Shanxi Province Applicant after: STATE GRID ELECTRIC POWER Research Institute OF SEPC Address before: 030024 No. 6, Qingnian Road, Datong City, Shanxi Province Applicant before: STATE GRID ELECTRIC POWER Research Institute OF SEPC |
|
| CB02 | Change of applicant information |