Network Decomposition For OPF Learning

Abstract

From the literature, it is observed that the machine learning techniques can pro vide quick and adaptable training time for the prediction of AC-OPF solution. To achieve this, a two-stage methodology is presented that takes benefit of a dimensional power network decomposition into a few sets of small areas. The first stage learns the voltages on the buses and power flows on the coupling lines of the areas. The second-stage model utilizes the information from the first stage to train the remaining parameters of the areas. This method can be demonstrated on the power system net works IEEE 14, 118 bus and Northern Regional Power Grid (NRPG) of Power Grid Corporation of India Limited (PGCIL) having 246 buses. This method can forecast AC-OPF solutions with accuracy and it can be predicted within small training time. Thus, it extends the possibility for machine-learning techniques to respond quickly to changes in the operating conditions. In order to reduce training time further the phasor measurement units (PMUs) are used in this thesis which provides real-time data for operation and control.