Research And Design Of Topology Information Mining System Based On Big Data Platform

With the rapid development and extensive applications of big data,different data types and application scenarios present a variety of new requirements for current data mining technologies, including time series and graph mining. Time series can reflect the dynamic characteristics of state parameters. Graph mining can get the structural characteristics of the network.However, for various reasons, the time series and the graph mining technology are not well applied. In real industrial production scenarios such as power network, the state parameters of topology nodes are collected as time series,with which the grid structure can be characterized.The main content of the dissertation is to deal with the time series pretreatment and graph mining algorithms corresponding to the electric power network operating scene design, and realize it through big data platform deployment. The dissertation focuses on the integration of time series and graph mining algorithms, and explores the technical details including time. series segmentation technique, time series similarity measure, graph influence propagation model algorithm and topology clustering algorithm, and use the analysis of the actual grid data to integrate and apply above key technology.Based on the existing time series preprocessing analysis method,characteristics of data fluctuation and the merits of traditional segmentation algorithm, the monotonic equidistant algorithm of time series is designed and realized. The similarity degree of production scene data is found in segmented structure. The similarity measure of dynamic influence is designed for segmented sequences at the time of state mutation. Based on the above-mentioned time series pretreatment steps and the characteristics of the connection structure of the grid topology, the influence degree of the node state similarity coefficient is measured by the abrupt change of the node state similarity coefficient. The algorithm of the topology influence propagation based on the segmentation mutation is proposed, and the Fast Unfolding is applied to Grid Topology cluster analysis.Finally, in order to build the grid topology information mining system, the dissertation constructs the Spark data processing platform and completes the Spark application module development of the above-mentioned time series preprocessing and graph mining algorithm, and then tests and analyzes the feasibility of the above algorithm on cluster platform. The end of the dissertation summarizes the main contents of this project, and further improvement and research direction are put forward according to the shortcomings.