Online Topology Identification For Smart Distribution Grids

Smart distribution grid is one of the key compositions of smart grid.To improve the security and economy of the power grid,flexible and frequent topology reconfiguration has become a basic characteristic of smart distribution grids.Most functions of the distribution management system(DMS),such as state estimation,power flow calculation,and voltage control,require the current topology of the grid.Therefore,it is of great significance to exploit methods for topology identification which are applicable to smart distribution grids with high penetration of distributed energy resources.According to the characteristics and operation status of distribution grids,a framework for topology identification of smart distribution grids based on machine learning is developed in this thesis.Under the framework,the machine learning method can be trained offline and applied online.There are two key issues should be addressed in the proposed framework.First,where and what type of measurements are required to guarantee the topology identification accuracy.Second,how to precisely fit the mapping relationship between input measurements and topologies.To solve these issues,the following researches are carried out in this thesis:Firstly,a feature selection method based on Light GBM is proposed.The importance of each feature can be obtained during the generation of the Light GBM,and based on which the most effective features for topology identification are selected.This process can greatly reduce the demand on the measurements in the network while keeps the accuracy of topology identification,which improve the feasibility and practicability of topology identification.Secondly,an online method for topology identification based on deep neural networks(DNN)is discussed.A deep neural network is constructed utilizing the nodal measurement snapshots.DNNs combine low-level features to form more abstract highlevel representations in order to discover the deep correlation of the data,and learn the mapping relationship between input measurements and corresponding topologies.After well-trained,a DNN can accurately identify the operational topology of the network with new measurement data.Considering the possibility of missing measurement data in practice,an imputation method based on the minimum variance is designed.The training sample which is most similar to the test sample can be found by calculating the variation of different voltage distribution curves,and the missing measurement data is estimated based on the most similar sample.Lastly,the proposed method are validated by IEEE 33-bus and PG&E 69-bus distribution grids,and sensitivity to different noise level and missing measurement data are further analyzed.The results indicate that the proposed method meets the demand of online applications with high accuracy,and is robust to noise and loss of measurement data.