Publications

Description

Authors: Adamantios Bampoulas, Fabiano Pallonetto, Eleni Mangina, Donal P. Finn

Abstract: This paper addresses the challenge of assessing uncertainty in energy flexibility predictions, which is a significant open question in the energy flexibility assessment field. To address this challenge, a methodology that quantifies the flexibility of multiple thermal and (opens in a new window)electrical systems is developed using appropriate indicators and considers the different types of uncertainty associated with building energy use. A Bayesian (opens in a new window)convolutional neural network is developed to capture aleatoric and (opens in a new window)epistemic uncertainty related to energy conversion device operation and temperature deviations resulting from exploiting building flexibility. The developed prediction models utilise residential occupancy patterns and a sliding window technique and are periodically updated. The (opens in a new window)energy systems evaluated include a heat pump, a (opens in a new window)photovoltaic system, and a stationary (opens in a new window)battery, and use synthetic datasets obtained from a calibrated physics-based model of an all-electric residential building for two occupancy profiles. Simulation results indicate that building flexibility potential predictability is influenced by weather conditions and/or occupant behaviour. Furthermore, the day-ahead and hour-ahead prediction models show excellent performance for both occupancy profiles, achieving coefficients of determination between 0.93 and 0.99. This methodology can enable electricity (opens in a new window)aggregators to evaluate building portfolios, considering uncertainty and multi-step predictions, to shift electricity demand to off-peak periods or periods of excess onsite (opens in a new window)renewable electricity generation in an end-user-customised manner.