Concerns about climate change, rising fossil fuel prices and energy security have spurred interest in renewable energy generation and smart grid. Due to the dynamic power demands and intermittent renewable energy resources, optimal scheduling of power generation systems is important to minimize cost and green house gas emissions, and to avoid blackouts in smart grid. In this paper, we propose a distributed stochastic scheduling scheme in smart grid communications with dynamic power demands and intermittent renewable energy resources. Due to meteorological instability and complex system dynamics, hidden Markov models are used in modeling renewable energy resources. We formulate the stochastic scheduling problem as a partially observable Markov decision process multi-armed bandit problem. A value iteration algorithm is used to solve the above problem. Simulation results are presented to show the effectiveness of the proposed scheme.