In order to enhance the searching ability of the flower pollination algorithm (FPA), this paper presents an improved flower pollination algorithm (IFPA) with three strategies, i.e., a double-direction learning strategy to advance the local searching ability, an imitative osphresis search strategy to strengthen the diversity of population and global searching ability, and a dynamic switching probability strategy to balance the switch between global and local searching. On this basis, a hydrothermal scheduling model with demand respond is proposed, which minimizes the fuel cost, maximizes the benefit of power supply company and reduces the electricity demand by providing an appropriate compensation to customers. Finally, the presented model is solved by using the IFPA. The simulation results show that the IFPA has outstanding performance, such as fast convergence speed and high accuracy, and the proposed model incorporating demand response can reduce the demand for electrical supply of customers and fuel cost of thermal power plants.