EOLD: A reinforcement learning-based energy-optimised load disaggregation framework for demand-side energy management

Abstract

Demand-Side energy Management (DSM) is a crucial strategy for balancing electricity supply and demand while enhancing energy efficiency, relying on sufficient data on electricity usage. Non-Intrusive Load Monitoring (NILM) is widely used for DSM strategies, as it effectively identifies the energy consumption of individual devices by measuring total power, significantly enhancing visibility. NILM should prioritise the dynamics of sub-load characteristics under future energy optimisation strategies rather than just historical data. For efficient load disaggregation, it must focus on optimising energy strategies. This study introduces a Reinforcement Learning-based Energy-Optimised Load Disaggregation (EOLD) framework to address this gap. The framework uses load disaggregation for final energy optimisation rather than initial sub-load characteristics. It utilises Reinforcement Learning (RL) to tackle the load disaggregation, with rewards focused on efficient, flexible, or economic energy goals. The Proximal Policy Optimisation (PPO) effectively disaggregates the air-conditioning load of three buildings, demonstrating the capabilities of the EOLD framework in optimising DSM for energy storage systems. The results show the proposed method optimises power curve flattening. It establishes a precise relationship between the main system's design power and the energy storage system's capacity. The framework can also be extended to disaggregate other flexible loads, such as photovoltaics and electric vehicles.