Data-Driven Modelling and Analysis in Power Distribution Automation

Abstract

The modern distribution networks (DNs) exhibit bi-directional power flows as opposed to the traditional one-way power flow, which has altered load demand patterns and behaviours. These changes make theDNs monitoring challenging, especially, with the presence of distributed en- ergy resources (DERs) locatedbehind the meter (BTM). Consequently, emerging load and gener- ation patterns are difficult to handle through standard state estimators and traditional methods of generating pseudo measurements. Thus, two data-driven approaches are proposed to improve the calculation of pseudo measurements and to improve the performance of forecasting-aided state estimation. Additionally, the modeling of high-impact devices influencing demand loads in modern DNs such as electric vehicles (EVs), heating, ventilation, and air conditioning sys- tems (HVAC), home battery energy storage systems (BESS), and rooftop photovoltaic (PV) are included in this thesis. The accuracy of the pseudo-measurement generation depends on the characteristics of the load demands such as the level of irregularities and the peaks. Therefore, a model that combines the strengths of WaveNet and LSTM architectures is proposed, as it can capture complex patterns of the power demand profiles with high levels of irregularities. While WaveNet can capture complex patterns in data, LSTM is effective in capturing temporal dependencies that are imposed by the high variability of the non-conventional power demand profiles. The challenges posed by the high variability of BTM DERs, due to the weather conditions, and potential failures or delays in the information and communication technology (ICT) system, are addressed by proposing a framework that includes constructing dynamic equivalent model (DEMs) for each node in the DNs. The DEMs are then integrated into Unscented Kalman Filter (UKF) to generate pseudo measurements and refine the real-time measurements.