Abstract: The economic dispatch problem (EDP) of power systems is approximated and converted into a differentiate one by way of maximum entropy in accordance to its characteristics. The different advantages and characteristics of conventional optimization methods and genetic algorithm are analyzed, then they are integrated together with a local searching operator introduced into implement rapid searching. An improved genetic algorithm is thus presented to solve EDP of power systems, of which the searching speed and local searching capability are improved by BFGS operator. At the same time, the uniform multiple individuals crossover operator is applied to improve global convergence and lead the population to the global optimum. Two examples are given to verify the validity of the algorithm.