A two-stage adaptive robust optimization model for power restoration in distribution networks with high penetration of renewable energy under extreme disasters is proposed in this paper. Uncertainty sets and adjustable robust parameters are employed to depict the uncertainty of renewable energy output and load demand. In the pre-disaster stage, unit commitment strategy and dispatch strategy of controllable generators are obtained to guarantee the reasonable distribution of power flow. In the post-disaster stage, network reconfiguration, emergency resources dispatch, adjustment strategy of controllable generators and load shedding are employed to perform fault recovery on the distribution network. The column and constraint generation algorithm (C&CG) is used to decompose the model into the main problem and subproblem. The dual theory and the big M method are employed to dualize and linearize the subproblem. The optimal recovery strategy can be obtained by alternating iterations between the main problem and the transformed subproblem. Case studies conducted on the improved PG&E 69-node system indicate that the proposed model is able to balance the robustness and economy under extreme disaster scenarios.