Effects of prediction errors on CO2 emissions in residential smart energy management systems with hybrid thermal-electric storage

Modern residential smart energy management systems allow for more efficient use of renewable energy through the application of various data-driven control strategies. Such strategies typically rely on predicting renewable power generation, domestic power demand, energy price and grid CO2 index. While the generation of such forecasts is well-researched, the impact of the associated prediction errors remains understudied.

This manuscript presents a generalised study of the effect of forecast errors on smart energy system performance. Results are obtained using multiple control optimisation techniques and real life data from residential dwellings spanning over multiple seasons.

Our analysis reveals that ideal forecasts can achieve up to 71.3% CO2 emissions savings compared to a baseline house without a smart energy system. The most significant performance decrease was caused by time lags in all three forecasts (grid CO2 index, solar power generation, and power demand). Among these, the CO2 index forecast was the most sensitive to errors, with an average performance deterioration of approximately 5% per 30 min of time lag. In contrast, errors in solar power generation and power demand forecasts had less impact, causing performance decreases of 18% and 21%, respectively, for extreme changes in forecast profile scale. This research identifies critical points in smart energy system design and offers insights to prioritise improvements in forecast models.