Machine learning based enhancements of security constrained optimal power flow computations


CALL: 2019

DOMAIN: MS - Materials, Physics and Engineering


LAST NAME: Capitanescu




KEYWORDS: sustainable energy systems, mathematical optimization, machine learning, data science, reliability, renewable energy, optimal power flow, reactive power reserves

START: 2020-05-01



Submitted Abstract

During the “energy transition” towards 100% clean energy supply, a key societal challenge, power systems are hosting steadily growing amounts of renewable energy sources (RES), which are gradually displacing large fossil-fuel power plants, and pose major challenges to reliable system operation. To handle these challenges requires re-thinking the methods adopted in existing optimal decision-making tools, by enriching them with suitable knowledge advances in other disciplines. This project is concerned with the key challenge of maintaining the transmission power grid reliability, which is under the responsibility of transmission system operator (TSO). Our specific objective is to significantly enhance the SCOPF algorithms under the Alternative Current (AC) physical model. From mathematical perspective, this AC-SCOPF problem is computationally hard (due to intrinsic challenging features like huge size, non-linearity and non-convexity). The limitations of existing algorithms to solve the AC-SCOPF problem currently prevent agile/informed TSO decision-making under increasingly variable operating conditions and hamper thereby cost-efficient and reliable integration of RES. The underlying hypothesis of this project is that machine learning (ML) methods can provide valuable insights (specifically computationally light proxy models of high precision) that may be plugged into various SCOPF-type optimization problems and vastly enhance AC-SCOPF solution algorithms. Based on this hypothesis the project addresses the two following research questions: Q1: How to significantly boost standard AC-SCOPF solution algorithms in terms of computational efficiency and interpretability? Q2: How to use these algorithms in order to identify and manage reactive power reserves (RPR) scarcity resulting from the energy transition? While both questions target computational advancement in solving hard optimization problems, Q2 further addresses an unexplored operation and knowledge gap. Indeed, RPR shortage is an anticipated new issue caused by the ongoing energy transition which affects directly the grid reliability and performance. To address these two research questions the project sets three overarching objectives: O1: to leverage machine learning in order to build proxy models that may effectively be exploited by existing optimisation solvers applied to the SCOPF formulations of Q1 and Q2.O2: to develop and validate a SCOPF solution algorithm enhanced with these ML-based proxy models that could be used close to real-time power system operation.O3: to develop a framework to identify and optimally manage RPR shortage via extended SCOPF formulation enhanced with ML-based proxy models. The methods and algorithms will be tested and validated using two benchmark power system models: a synthetic small case (IEEE-RTS96) of around 100 buses and 100 contingencies as well as a publicly available large real power system model of around 2,000 buses and 3,000 contingencies.

This site uses cookies. By continuing to use this site, you agree to the use of cookies for analytics purposes. Find out more in our Privacy Statement