An On-line Generator Start-up Strategy Based on Deep Learning and Tree Search
Runjia Sun, Yutian Liu · 2018
Considering the uncertainty of initial power network topology, restoration time of lines and downtime of generators during generator start-up, this paper presents an on-line decision making strategy of generator start-up based on deep learning and Monte Carlo tree search (MCTS). At first, in order to generate labeled samples of generator start-up atomically, a self-generated samples method is proposed. The sparse autoencoder (SAE) is applied to train the samples to establish a value network, which is used to estimate the optimal value of decision indicator under certain circumstances. Then, the upper confidence bound apply to tree (UCT) algorithm, move pruning technology and value network are applied to MCTS to search the line needed to be restored based on the situation of power system. Finally, a weighted total power generation capability is proposed to integrate the parallel computing results of MCTS and determine the next restored line. The New England 10-unit 39-bus power system is used to show the feasibility and effectiveness of the proposed strategy, and the on-line decision making strategy is compared with traditional method to show its effectiveness for uncertain situations in generator start-up.