Real-Time Price Elasticity Reinforcement Learning for Low Carbon Energy Hub Scheduling Based on Conditional Random Field

Abstract

Energy hub scheduling plays a vital role in optimally integrating multiple energy vectors, e.g., electricity and gas, to meet both heat and electricity demand. A scalable scheduling model is needed to adapt to various energy sources and operating conditions. This paper proposes a conditional random field (CRF) method to analyse the intrinsic characteristics of energy hub scheduling problems. Building on these characteristics, a reinforcement learning (RL) model is designed to strategically schedule power and natural gas exchanges as well as the energy dispatch of energy hub. Case studies are performed by using real-time digital simulator that enables dynamic interactions between scheduling decisions and operating conditions. Simulation results show that the CRF-based RL method can approach the theoretical optimal scheduling solution after 50 days training. Scheduling decisions are particularly more dependent on received price information during peak-demand period. The proposed method can reduce 9.76% of operating cost and 1.388 ton of carbon emissions per day, respectively. Index Terms-Energy hub, price elasticity, real-time digital simulator (RTDS), reinforcement learning (RL), conditional random field, carbon reduction.

Operation and Planning of Energy Hubs Under Uncertainty—A Review of Mathematical Optimization Approaches

“…Machine learning methods have been applied to both optimal operation [175], [176], [177] and optimal planning [178], [179] of EHs. They have a huge potential to improve the EHs' energy management, particularly when real-time control is needed [180], [181], [182], [183]. A distributed energy management approach based on multi-agent reinforcement learning has been applied to residential EHs to minimize the operation cost of EHs [184], [185].…”