Research On Distribution System State Estimation Based On Gaussian Process

With the advancement of sustainable development,a large number of distributed generations and electric vehicles are connected to power grid,which has brought many challenges to the operation and reliability of distribution system.Distribution system state estimation（DSSE）can help operators accurately perceive the operating state of power system and provide data support for subsequent optimization control,which is of great significance.This thesis,funded by the national key research and development program “Research on Key Technologies for Planning and Stability Control of Weakly Interconnected Hybrid Renewable Energy System”（2018YFE0127600）,conducts related researches about DSSE.The contents are divided into three parts as follows.First,there are only a small number of real-time measurement devices in real distribution system because of the lack of sufficient investment and technical support.In order to solve the problem of poor observability caused by insufficient measurement devices,this thesis proposes a pseudo-measurement construction method based on net load forecasting.The proposed method takes historical characteristics of net load and time information as input information,and then generates reliable node power pseudo-measurement data by learning the mapping relationship between input information and net load.The numerical experiments on a distributed energy micro-grid in Australia show that the proposed method obtains better results than other net load forecasting methods and can control the error of pseudo-measurement data within 10%.Second,considering the problem that traditional physical model-based DSSE methods are limited to apply in practice because it is difficult to obtain accurate network topology and line parameters of real distribution system,this thesis proposes a data-driven DSSE model considering pseudo-measurement data.Combined with the pseudo-measurement construction method proposed above,the proposed model can obtain state estimation results using a small amount of branch power real-time measurement data and a large number of node power pseudo-measurement data.The numerical experiments on IEEE 33-bus and 119-bus distribution systems show that after training with massive historical data in the supervisory control and data acquisition（SCADA）,the proposed model can obtain more accurate deterministic estimation results and more reliable interval estimation results than traditional physical models and other data-driven models without the distribution system topology and line parameters.Finally,the network topology of distribution system changes frequently in practice,which causes the new topology to be unknown and the historical data under new topology to be insufficuient.In order to solve the problem that it is difficult to obtain reliable estimation results for data-driven DSSE models when the training data is insufficient in new topology,this thesis proposes a data-driven DSSE model considering topology transformation.By constructing a multi-task learning structure between the DSSE under new topology and the DSSE under other topology,the proposed model can learn the correlation between different topologies and utlizes it to improve the performance of data-driven DSSE under new topology.The numerical experiments in different distribution systems demonstrate that the proposed model can obtain excellent voltage estimation results when the new topology has only a small amount of training data.