A coordinated beamforming design for heterogeneous wireless networks combining radio frequency (RF) and visible light communication (VLC) access points (AP), with several users receiving information simultaneously from both APs, was investigated. Our goal was to maximize the energy efficiency (EE) of the entire system under both the perfect and imperfect channel state information (CSI) conditions. For such optimization, we formulated a fractional programming problem, and to overcome its non-convex objective function and constraints, we developed four successive convex approximation methods. Our extensive numerical experiments demonstrated that these algorithms can achieve near-optimal performance. Furthermore, in the perfect CSI case, EE performance was improved by $\text{62}{\%}$ over the maximum ratio transmission (MRT) scheme, and EE of the proposed RF/VLC network increased by $\text{41}{\%}$ compared with traditional that of the RF heterogeneous wireless networks. In the imperfect CSI case, we first investigated the importance of robust beamforming when channel errors occur. Moreover, the proposed algorithm and network architecture outperformed the MRT scheme and the traditional heterogeneous wireless network by $\text{229}{\%}$ and $\text{93}{\%}$ , respectively. In summary, the careful design of an energy-efficient beamforming scheme was proved essential, and the proposed mixed RF/VLC network architecture was much more energy efficient than those without VLC.