Due to its high mobility and low cost, unmanned aerial vehicle mounted base station (UAV-BS) can be deployed in a fast and cost-efficient manner for providing wireless services in areas where traditional terrestrial infrastructures cannot be laid for technical and economic reasons. In this letter, we investigate the problem of joint three-dimensional (3D) deployment and power allocation for maximizing the system throughput in a UAV-BS system. To solve this non-convex problem, we propose a deep deterministic policy gradient (DDPG) based algorithm. The proposed algorithm allows the UAV-BS to explore in continuous state and action spaces to learn the optimal 3D hovering location and power allocation. Simulation results show that the proposed algorithm outperforms the traditional deep Q-learning-based method and genetic algorithm.