For the uncertainty of the initial condition of power system after blackout and the restoration condition of transmission lines during restoration

an online decision-making strategy for generator start-up is proposed based on deep learning and Monte Carlo tree search(MCTS). A method for training samples is proposed to generate samples of generator start-up atomically. The sparse autoencoder(SAE)is introduced to train the training samples for a value network. Based on the condition of power system

the modified upper confidence bound apply to tree(MUCT)algorithm

value network and move pruning are applied in MCTS to search the next restored line. The weighted total generation capability is proposed as a decision-making index to summarize the parallel computing results of MCTS and decide the next restored lines. Results of IEEE New England 10-unit 39-bus power system and the Western Shandong power grid demonstrate the feasibility and effectiveness of the proposed strategy

and the online strategy is compared with traditional strategy to show its robustness and ability of handling different uncertain situations during generator start-up.