This paper proposes a decentralized collaborative control framework of autonomous VGT (virtual generation tribe) for smart grid. A VGT-CCA (VGT based collaborative consensus algorithm) is firstly developed to solve the dynamic GCD (generation command dispatch) optimization of the AGC (automatic generation control) under an ideal communication network. Then a novel CCA VGT-RCCA (VGT based robust CCA) is designed by introducing the consensus gain functions and virtual consensus variables, which provides significant robustness to a practical communication network consisted with switching topology, transmission delay and noise. The performance of VGT-CCA and VGT-RCCA has been evaluated on a typical two-area load frequency control model and the China southern power grid model, respectively. Simulation results verify the effectiveness of the proposed algorithms. (C) 2016 Elsevier Ltd. All rights reserved.