Electricity demand flexibility estimation in warehouses using machine learning

In this chapter, the applications of providing flexibility in buildings and the corresponding commonly employed methodologies are first reviewed. Subsequently, physics-based simulations are performed to model the behavior of a warehouse that is undergoing flexibility events, enabled through the cooling system’s setpoint adjustments. The resulting data are employed to train machine learning (ML)-based pipelines that predict the event’s duration. Accordingly, linear regression and six different ML algorithms are trained using the obtained features, and their estimation performance is compared. The best-performing pipeline is then employed in a feature selection process to find the most relevant features. Results indicate that employing the identified most promising algorithm (random forests) permits predicting the flexibility duration with an average mean absolute percentage error of 6.75%, providing a suitable tool for planning participation in demand response programs or charging electric vehicles.