Study On Clustering And Neural Network Recognition Of Icing Events On Transmission Lines

Loads induced by atmospheric icing on transmission lines may lead to ice flashover,conductor galloping,even conductor breakage,tower collapse,etc.,seriously threatening the reliability of the power supply and integrity of the power transmission infrastructure.The term 'icing events' used in this paper refers to the entire process from the appearance to the disappearance of ice accretion on the transmission lines,and the types of icing events can reveal the changing law of the ice thickness during this process.The research on types of icing events makes sense to the anti-icing deployment,emergency strategies,ice thickness forcasting,and anti-icing design of transmission lines.However,less attention is focused on the types of icing events at home and abroad.There are also some problems of concern,including the lack of systematic summary,classification methods and recognition methods of the types of icing events.With the developments and popularization of icing monitoring technology,massive icing data make it feasible to study the types of icing events on transmission lines.Meanwhile,artificial intelligence algorithms have also been widely applied in the field of ice accretion.Therefore,the dissertation is mainly focused on the types of icing events on transmission lines in the China Southern Power Grid based on the monitoring data and artificial intelligence algorithms,which aims to comprehensively summarize the types and appropriately propose the recognition methods.Firstly,two mechanical calculation models of ice thickness for the suspension tower and the tensile tower are optimized,and the 97 independent icing events are collected from the ice thickness data transferd by the mechanical data.Then,a clustering method for icing events is considered to study the types of icing events.In order to solve the problems of unequal-length and large-scale sequences of different icing events,a hierarchical,11 characteristic parameters that can characterize the ice thickness evolution curves are presented.For the problem of uncertain number of types,a hierarchical weighted K-means clustering method is proposed.The clustering results of 97 icing events show that there are 6 types of icing events in the China Southern Power Grid.Finally,a recognition method for the types of icing events based on the artificial neural network is proposed.After analyzing the information between ice accretion and different micro-meteorological parameters based on the maximum information coefficient theory,the temperature is selected as the best feature for neural network.The BP neural network and GRNN neural network are constructed and trained separately.The prediction results show that the accuracy of GRNN for icing events recognition is 70% ? 80%,which meets the accuracy requirements.The work of this dissertation will contribute to further understanding of the icing events in the Southern Power Grid,which has high reference value for other research of ice accretion on transmission line.