CN-122020247-A - Power grid fault prediction method and system based on power grid fusion terminal

CN122020247ACN 122020247 ACN122020247 ACN 122020247ACN-122020247-A

Abstract

The invention discloses a power grid fault prediction method and a power grid fault prediction system based on a power grid fusion terminal, which are used for obtaining a data feature set adapting to a current version based on a data set of a historical version and a data feature space of the current version of the power grid fusion terminal, obtaining a migration control strategy based on the data feature set adapting to the current version and a version upgrading strategy, solving the problem of data version compatibility of the power grid fusion terminal, fusing static attribute data and dynamic time sequence data into a space-time weight graph based on a power grid topology, predicting faults of a power grid based on the space-time weight graph, and solving the problem of deep coupling of dynamic and static data.

Inventors

LIU QIANG
ZHANG RUIFANG
ZHAO WENYI
LI JIAKANG
YANG YUNQI
ZHANG CHENG
REN RUIJIE
LUO XIAOHAN
LIANG YULONG
MA JIA

Assignees

国网陕西省电力有限公司铜川供电公司

Dates

Publication Date: 20260512
Application Date: 20260127

Claims (10)

1. The power grid fault prediction method based on the power grid fusion terminal is characterized by comprising the following steps of: Acquiring a data feature set adapting to a current version based on a data set of a historical version of a power grid fusion terminal and a data feature space of the current version, and acquiring a migration control strategy based on the data feature set adapting to the current version and a version upgrading strategy; Dividing the data feature set adapting to the current version into static attribute data and dynamic time sequence data; Fusing the static attribute data and the dynamic time sequence data into a space-time weight graph based on the power grid topology; And carrying out fault prediction on the power grid based on the space-time weight graph.
2. The power grid fault prediction method based on the power grid fusion terminal according to claim 1, wherein the power grid fault prediction method is characterized by comprising the following steps of: the obtaining the data feature set adapting to the current version based on the data feature set of the historical version and the data feature space of the current version of the power grid fusion terminal, and obtaining the migration control strategy based on the data feature set adapting to the current version and the version upgrading strategy comprises the following steps: acquiring semantic difference reports of a historical version and a current version based on feature vectors of a data set of the historical version of the power grid fusion terminal and a data feature space of the current version; And processing the feature vector of the data set of the historical version based on the semantic difference report, so that the feature vector of the data set of the historical version is mapped to the data feature space of the current version and has the same dimension as the data feature space of the current version.
3. The power grid fault prediction method based on the power grid fusion terminal according to claim 2, wherein the power grid fault prediction method is characterized by comprising the following steps of: The semantic difference report of the historical version and the current version is obtained based on the feature vector of the data set of the historical version of the power grid fusion terminal and the data feature space of the current version, and the semantic difference report comprises the following steps: Training a comparison learning network through the feature vector of the data set of the historical version and the feature vector of the data feature space of the current version, and calculating the similarity of the feature vector of the data set of the historical version and the feature vector of the data feature space of the current version through the trained comparison learning network to obtain a similarity matrix; And acquiring semantic drift characteristics and dimension change characteristics of the characteristic vectors of the data set of the historical version and the characteristic vectors of the data characteristic space of the current version based on the similarity matrix.
4. The power grid fault prediction method based on the power grid fusion terminal according to claim 2, wherein the power grid fault prediction method is characterized by comprising the following steps of: The processing the feature vector of the data set of the historical version based on the semantic difference report, so that the feature vector of the data set of the historical version is mapped to the data feature space of the current version and has the same dimension as the data feature space of the current version, includes: Training a feature mapping model by reinforcement learning, and mapping the semantic drift features to a data feature space of a current version with the aim of minimizing reconstruction errors in the feature space of the current version; Expanding the dimension of the dimension change feature by generating an countermeasure network to be the same as the dimension of the data feature space of the current version; Constructing a strategy migration engine based on reinforcement learning, and obtaining an adaptation strategy through the current version characteristics of the historical strategy execution result; and fine-tuning the adaptive strategy parameters through transfer learning.
5. The power grid fault prediction method based on the power grid fusion terminal according to claim 1, wherein the power grid fault prediction method is characterized by comprising the following steps of: the fusing the static attribute data and the dynamic time sequence data into a space-time weight graph based on the power grid topology comprises the following steps: A static data feature vector obtained by encoding the static attribute data; Extracting time sequence characteristics of the dynamic time sequence data through a long-short-period memory network, counting statistical characteristics of the dynamic time sequence data through a sliding window, and splicing the time sequence characteristics and the statistical characteristics to form a dynamic data characteristic vector; based on the power grid topology, taking a power grid fusion terminal as a node, taking the static data feature vector as a node initial attribute, setting spatial weight and dynamic weight, fusing the spatial weight and the dynamic weight, and constructing a space-time weight map.
6. The power grid fault prediction method based on the power grid fusion terminal according to claim 1, wherein the power grid fault prediction method is characterized by comprising the following steps of: the performing fault prediction on the power grid based on the space-time weight graph comprises the following steps: transmitting node characteristics through a graph neural network, and aggregating neighbor information; the attention mechanism dynamically adjusts the importance of the nodes and outputs fault probability prediction; And positioning fault equipment based on the edge weights and the node characteristics of the space-time weight graph.
7. The power grid fault prediction method based on the power grid fusion terminal as claimed in claim 4, wherein: the constructing a policy migration engine based on reinforcement learning, obtaining an adaptation policy through the current version characteristic of the history policy execution result, including: Combining the historical dimension and the newly added dimension of the data feature set adapting to the current version, constructing a new state space, mapping the data feature set adapting to the current version to the new state space through a lightweight mapping network, and constructing an action space based on the data feature set adapting to the current version; Reinforcement learning is based on a dual-branch structure output historical state action distribution and a new state action distribution, and a strategy gradient method is used for updating network parameters so as to maximize a jackpot as an optimization target for training.
8. The power grid fault prediction method based on the power grid fusion terminal according to claim 5, wherein the power grid fault prediction method is characterized by: The static data feature vector obtained by encoding the static attribute data comprises: performing single-heat coding on the static attribute data; Traversing the static attribute data, counting all unique categories, constructing a category set, and determining the dimension of the category set; Constructing a coding matrix based on the set of categories; Normalizing the continuous attribute data in the coding matrix; and generating a static data characteristic vector through splicing.
9. The power grid fault prediction method based on the power grid fusion terminal according to claim 5, wherein the power grid fault prediction method is characterized by: Based on the power grid topology, taking a power grid fusion terminal as a node, taking the static data feature vector as a node initial attribute, setting space weight and dynamic weight, fusing the space weight and the dynamic weight, and constructing a space-time weight map, wherein the space-time weight map comprises the following steps: Performing single-heat coding or embedding on discrete static features, normalizing continuous features, and splicing all the features; calculating an initial side weight construction space weight based on the equipment space distance; quantifying the similarity of dynamic characteristics by a dynamic time warping algorithm, and dynamically adjusting the side weight to obtain dynamic weight; and linearly weighting and fusing the space weight and the dynamic weight to construct a space-time weight map.
10. A power grid fault prediction system based on a power grid fusion terminal is characterized by comprising: the cross-version data processing module is used for obtaining a data feature set adapting to the current version based on a data set of a historical version of the power grid fusion terminal and a data feature space of the current version, and obtaining a migration control strategy based on the data feature set adapting to the current version and a version upgrading strategy; the data classification module is used for dividing the data feature set adapting to the current version into static attribute data and dynamic time sequence data; the data fusion module is used for fusing the static attribute data and the dynamic time sequence data into a space-time weight graph based on the power grid topology; And the fault prediction module is used for predicting the faults of the power grid based on the space-time weight map.

Description

Power grid fault prediction method and system based on power grid fusion terminal Technical Field The invention relates to the technical field of power systems, in particular to a power grid fault prediction method and system based on a power grid fusion terminal. Background Along with the accelerated evolution of the smart power grid to digitization and intellectualization, the power grid fusion terminal is used as a core node for data acquisition and control, and frequent iteration of software versions is needed to adapt to novel equipment access and function upgrading. However, version iteration causes mismatch of historical data and new version feature space, dynamic and static data are difficult to deeply fuse due to space-time isomerism, and accordingly failure prediction accuracy is lowered, and control strategies are invalid. The traditional method depends on a static characteristic mapping table or a fixed weight knowledge graph, cannot dynamically adapt to the topological change of the power grid and the semantic drift of data, and cannot meet the severe requirements of a novel power system on instantaneity, accuracy and expandability. Therefore, a collaborative architecture with cross-version compatibility and dynamic and static data deep coupling is constructed, and the collaborative architecture becomes a key for breaking through the technical bottleneck of the power grid fusion terminal. Disclosure of Invention In view of the above problems, the invention provides a power grid fault prediction method and a system based on a power grid fusion terminal, so as to solve the problems of cross-version compatibility and dynamic and static data deep coupling in the power grid fault prediction process. The technical scheme of the invention is as follows: a power grid fault prediction method based on a power grid fusion terminal, the method comprising: Acquiring a data feature set adapting to a current version based on a data set of a historical version of a power grid fusion terminal and a data feature space of the current version, and acquiring a migration control strategy based on the data feature set adapting to the current version and a version upgrading strategy; Dividing the data feature set adapting to the current version into static attribute data and dynamic time sequence data; Fusing the static attribute data and the dynamic time sequence data into a space-time weight graph based on the power grid topology; And carrying out fault prediction on the power grid based on the space-time weight graph. Further, the obtaining the data feature set adapted to the current version based on the historical version data set and the current version data feature space of the power grid fusion terminal, and obtaining the migration control policy based on the data feature set adapted to the current version and the version upgrading policy, includes: acquiring semantic difference reports of a historical version and a current version based on feature vectors of a data set of the historical version of the power grid fusion terminal and a data feature space of the current version; And processing the feature vector of the data set of the historical version based on the semantic difference report, so that the feature vector of the data set of the historical version is mapped to the data feature space of the current version and has the same dimension as the data feature space of the current version. Further, the semantic difference report of the historical version and the current version is obtained based on the feature vector of the data set of the historical version and the data feature space of the current version of the power grid fusion terminal, and the semantic difference report comprises the following steps: Training a comparison learning network through the feature vector of the data set of the historical version and the feature vector of the data feature space of the current version, and calculating the similarity of the feature vector of the data set of the historical version and the feature vector of the data feature space of the current version through the trained comparison learning network to obtain a similarity matrix; And acquiring semantic drift characteristics and dimension change characteristics of the characteristic vectors of the data set of the historical version and the characteristic vectors of the data characteristic space of the current version based on the similarity matrix. Further, the processing the feature vector of the data set of the historical version based on the semantic difference report, so that the feature vector of the data set of the historical version is mapped to the data feature space of the current version and has the same dimension as the data feature space of the current version, includes: Training a feature mapping model by reinforcement learning, and mapping the semantic drift features to a data feature space of a current version with the aim of minimizing reconstructi