Abstract: In order to seek the solutions of many discontinuous portfolio optimization models in stock market promptly and efficiently, an improved optimization algorithm was designed. By introducing crossover operations in genetic algorithm, an innovative particle swarm optimization algorithm (INPSO) based on optimal and suboptimal locations was proposed. And a performance test reveals that the algorithm performs better than some improved particle swarm optimization algorithms，and overcomes the problem of prematurity. Then in simulation experiment, by use of INPSO the optimal values of two portfolio models under different expected return rates were thus obtained, and the algorithm shows a marvelous convergence property in the iteration process.