The Smart Grid market is dynamic and complex, and brokers are widely introduced to better manage this market. This paper proposes an intelligent broker model—GongBroker, with smart trading strategies to cope with the dynamics and complexity. GongBroker first predicts the short-term demands of various consumers, and then buys energy from the wholesale market through auctions, and sells energy to various consumers in the retail market. In order to predict the customer demand, a data-driven method is proposed. All the consumers are hierarchically clustered according to their historical energy consumptions, and different prediction methods are tailored for different customer clusters to predict one-day-ahead hourly energy demand. Based on the predicted demand, the GongBroker employs a Markov Decision Process for the one-day-ahead auction in the wholesale market. To compete with other brokers, GongBroker uses independent reinforcement learning processes to optimize prices for different types of consumers. Experimental results demonstrate that GongBroker not only is competitive in making profit, but also keeps a good supply-demand balance.