Power system skeleton restoration is of multi objectives， multi variables， multi constraints and uncertainties， difficult to establish accurate mathematical models. It needs to comprehensively consider various constraints， such as transient overvoltage， sustained power frequency overvoltage， self-excitation， frequency fluctuation， generating unit and load characteristics， restoration duration， available transfer capability of transmission line， equipment failure probability， etc. Based on the data warehouse technology， a group intelligent decision support system model is constructed. It consists of three parts， i.e.， data platform， intelligent control sub-system and group decision support sub-system. A three-stage skeleton restoration strategy is proposed to control the restoration process. According to the evaluation value of each benefit-oriented attribute， the multi-attribute utility theory is employed to dynamically determine the next restoration target. Simulation results on a real power system demonstrate that this method can not only generate feasible cases for case-based reasoning， but also provide online supports in actual process of restoration.