Research On Risk Identification Model Of Power Grid Equipment Based On Distributed Big Data Flow Classification

Due to the high development of smart grids in recent years,the scale of power grids has been increasing and becoming more complex,which makes the maintenance and maintenance of power transmission and transformation equipment in the power grid more complicated.However,in order to meet the requirements of ensuring stable operation of the power grid,improving power quality,and reducing maintenance costs,it is necessary to further upgrade and improve the health status and working years of the power grid equipment.At the same time,the risk in the power system is composed of faults and abnormalities of the power grid.Therefore,the identification of the risk of the power grid equipment is equivalent to the accurate judgment of faults and abnormalities in the power grid equipment.However,with the large-scale application of information collection and transmission technology in the power system,the online monitoring system for each device in the power grid has been continuously improved,thus forming a large number of data streams implying abnormal information of the power grid equipment.In this paper,from the perspective of data stream processing,in order to solve the real-time detection problem for grid equipment risk,the following research is carried out.Aiming at the cleaning problem of grid equipment online monitoring data flow,this paper proposes a method for cleaning the grid data stream based on association rules.This method uses association rules to analyze the correlation strength of each sequence in the data stream,and proposes an anomaly data detection algorithm to identify the anomaly data,and combines the improved wavelet neural network to complete the cleaning of the data stream.According to the relationship between each sequence,combined with the abnormal data filtering algorithm based on sliding window to identify the bad data in the data stream,the data is cleaned by the improved wavelet neural network.Experiments show that this method can effectively identify bad data and fault data caused by equipment anomalies,and the improved wavelet neural network has better cleaning effect.In order to solve the real-time detection problem of grid equipment anomaly,this paper proposes a data stream classification model based on distributed processing based on the above data cleaning algorithm.In order to realize distributed processing of grid data stream,a local node mining method is designed,and based on uneven data.The global mining mode of flow classification.Therefore,in the fourth chapter,in this paper,in order to realize the classification of grid equipment data flow,it is necessary to construct a data stream classification model based on distributed processing,then select the corresponding data sequence and format abstraction,and design the local node mining method under this model.With global mining mode.A block-to-block mining strategy is implemented in the local node miner by acquiring the current data block.A block-to-block mining strategy is implemented in the local node miner by acquiring the current data block.At the same time,combined with the clustering algorithm to complete the expression and real-time maintenance of the local mining model,and thus improve the information transmission rate between the nodes,and ensure the timeliness of the overall classification algorithm.Aiming at the problem of unbalanced data distribution in each category of grid equipment data flow,this paper proposes an integrated classification algorithm based on unbalanced data stream in global miner.The method first receives the micro clusters transmitted by the respective local nodes at the central node.Then,in order to train and update the basic classifier in the integrated classifier,a micro-cluster-based learning sample restoration algorithm is proposed.The training data block is constructed by using the restored data and the abnormal data set after data cleaning,so as to improve the identification accuracy of the classifier to the risk category of the power grid equipment.A selective resampling mechanism is designed to balance the distribution of each category in the current data block,thereby avoiding the problem of inefficient identification of small categories of data.At the same time,in order to reduce the impact of concept drift on the accuracy of the classifier,periodic incremental updates are performed for the integrated classifier in the global mining mode,and the samples correctly predicted by the basic classifier are eliminated when the classifier is updated.,not using it to train other basic classifiers to improve the diversity of integrated classifications,and then to adapt to the concept drift.So far,the task of the global mining mode has been completed,and combined with the local mining mode,the construction of the distributed data-based big data stream classification mode is realized.To solve the problem of real-time identification of grid equipment risks.Finally,the performance of the classification model and the classification effect of the unbalanced data stream and the conceptual drift data stream are verified.The results verify the effectiveness of the proposed method.