In this seminar, Sor oush Karimi will give a talk about a Data-driven Trading Algorithm fo r Short-Term Electricity Market.
Growth in the penetration of renewable energy sources makes electricity supply more uncertain and leads to an increase in the grid system imbalance and imbalance ( real-time) price volatility. In some European countries, large-scale batteries can be used to support transmission system operators (TSOs) in maintaining grid stability and earn profit, a practice called imp licit balancing. Model predictive control (MPC) strategies to exploit these implicit balancing strategies capture arbitrage opportunities but fail to accurately model imbalance markets and face high computat ional costs. Model-free reinforcement learning (RL) methods are fast to execute but require data-intensive training and usually rely on re al-time and historical data for decision making. In this talk, I will present our proposed MPC-guided distributional RL method that combin es the complementary strengths of both MPC and RL. The proposed metho d can effectively incorporate forecasts into the decision making proc ess (as in MPC), while maintaining the fast inference capability of R L.
DESCRIPTION: In this seminar, Soroush Karimi will give a talk about a Data-driven Trading Algorithm for Short-Term Electricity Market. Grow th in the penetration of renewable energy sources makes electricity s upply more uncertain and leads to an increase in the grid system imba lance and imbalance (real-time) price volatility. In some European co untries, large-scale batteries can be used to support transmission sy stem operators (TSOs) in maintaining grid stability and earn profit, a practice called implicit balancing. Model predictive control (MPC) strategies to exploit these implicit balancing strategies capture arb itrage opportunities but fail to accurately model imbalance markets a nd face high computational costs. Model-free reinforcement learning ( RL) methods are fast to execute but require data-intensive training a nd usually rely on real-time and historical data for decision making. In this talk, I will present our proposed MPC-guided distributional RL method that combines the complementary strengths of both MPC and R L. The proposed method can effectively incorporate forecasts into the decision making process (as in MPC), while maintaining the fast infe rence capability of RL. END:VEVENT END:VCALENDAR