Machine learning applications in power flow analysis

In this project, we explore the viability of solving probabilistic load flow (PLF) based on Gaussian Process (GP) regression. The objective is to find solutions to the power flow problem to a power system operating under load uncertainty and make better operational decisions. GP Learning is used to...

Full description

Saved in:
Bibliographic Details
Main Author: Kyaw, Myo Thi Ha
Other Authors: Hung Dinh Nguyen
Format: Final Year Project
Language:English
Published: Nanyang Technological University 2021
Subjects:
Online Access:https://hdl.handle.net/10356/149777
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Nanyang Technological University
Language: English
Description
Summary:In this project, we explore the viability of solving probabilistic load flow (PLF) based on Gaussian Process (GP) regression. The objective is to find solutions to the power flow problem to a power system operating under load uncertainty and make better operational decisions. GP Learning is used to execute training and testing over probability distributions of power injections at load buses in IEEE 14-bus and 30-bus systems. To optimize the PLF approach, simulations were performed on the IEEE 14-bus and IEEE 30-bus system with control variables such as training data set size and standard deviation. GP Learning is targeted at reducing error margin between predicted values and simulated values from traditional power flow solvers, reducing computational time for large sample sizes as compared to Monte Carlo Simulation approach. The results performed on the power systems show that the proposed method can learn the nodal voltage GP function over the power injection distribution only using a small number of training samples. Testing also indicates that probabilistic load flow solutions with low error margin of order 10^-9 can be achieved on a scale of 10^5 test points.