Abstract
Customer baseline load (CBL) prediction plays an important role in calculating the volume and value of the flexibility provided by end-users. In this paper, two different CBL methods are applied to investigate their prediction accuracy for a given load with high resolution metered data. One of the CBL methods makes use of historical data, named CBLXofY , while the other makes use of the load pattern before/after the CBL-prediction, denoted as CBLB/A . A real office with high resolution load data is used to investigate CBL prediction accuracy at multiple measuring points, for the different CBL methods. The results show that CBLB/A has a high level of accuracy at the office-building level, due to an internal 200-kW threshold for import that made the load profile flat during the midday. As the load increases throughout the morning, both methods undershoot the accuracy, where CBLB/A undershoots by 13%, while CBLXofY undershoots by 4–5%. In the electric vehicle (EV) parking lot, there is a noticeable offset for both CBL-methods, as the lot is the internal throttling mechanism for maintaining the 200-kW threshold at the building level. This analysis has captured the importance of measuring points for calculating CBL when an internal demand response is available within the building, which can cause noise and inaccuracy.
