In traditional Light-Fidelity (Li-Fi) attocell networks, the dense structure of Li-Fi access points (APs) leads to severe inter-cell interference (ICI). User-centric (UC) networks, which dynamically reform amorphous cells as the user mobility varies in time, mitigate the effect brought by ICI. In this work, we propose a new user-centric clustering algorithm with the objective of maximizing energy efficiency (EE). In previous studies, clustering algorithms usually merely aim at minimizing the distance from the centroid of the cluster to UEs belonging to the cell. Since ICI is the main challenge in Li-Fi networks, separating cells as far as possible is also important. Therefore, we propose a new clustering algorithm that also considers the distance among constructed cells to yield better energy efficiency as the power consumption is further reduced.