Embedding Temporal Awareness in Reinforcement Learning Models for Energy System Control

A reinforcement learning (RL) agent can capture energy system dynamics from historical data to improve its decision-making.However, the RL agent struggles to fully understand temporal patterns in scenarios with high fluctuations.This study addresses this gap by embedding temporal awareness into RL agents for energy system control.The proposed approach integrates generation and load time series into the state representation using a temporal embedding module based on long short-term memory (LSTM).The method is evaluated on highly variable patterns from the Netherlands to demonstrate the performance of temporal RL decision-making in different conditions.