Abstract: This paper proposes two improved harmony search algorithm for solving numerical optimization problems.To get start,a roulette and self-adaptive harmony search is proposed,in which a roulette random choosing mechanism and a pitch self adaptive adjusting strategy is introduced to overcome defects of slow convergent speed and parameter settings difficulties respectively in current harmony search algorithms.Then a more effective harmony search memetic algorithm fused the improved harmony search as a global search with Powell's method as the meme unit is implemented.To jump out local optima,chaos disturbance is also employed.Simulation with six benchmark functions shows that the two improved harmony search algorithms improve harmony search algorithm effectively.