The power system economic dispatch problem is an important research topic in power systems. To solve this problem, an improved particle swarm optimization(ODPSO) algorithm is proposed. In the early stage of the improved algorithm, the generalized opposition-based learning strategy is used to make the algorithm quickly close to the potential search area and improve the convergence speed. In the later stage of searching, inspired by evolutionary process of differential evolution, an improved mutation and crossover strategy is developed to update the optimal particle of the current population, thus improving the population diversity and assisting the algorithm to obtain global optimal solution. In order to validate the effectiveness of the improved algorithm, 22 constraint test functions presented in CEC2006 are simulated. Experimental results show that the improved algorithm is superior to other compared algorithms in terms of the accuracy and stability. Finally, the improved algorithm is applied to two economic dispatch problems of power systems, which takes into account the ramp rate limits of the generating units, prohibited operating zones and power balance constraint, and satisfying results are obtained.