In view of the situation that original quantum particle swarm optimization ( QPSO) is prone to premature and has poor convergence accuracy in complex multimode functions, an improved QPSO algorithm with random disturbance mechanism (MQPSO) is proposed. In the design of the improved algorithm, the idea of cross operator in genetic algorithm and random disturbance operation are applied to locate each particle’ s history best position and global optimal position for enhancing algorithm convergence ability in later iterations and maintaining the diversity of population. Moreover, to improve algorithm ’ s global convergence rate and accuracy, the contraction-expansion factor, an important parameter in QPSO algorithm is adjusted nonlinearly. Through eight test functions, MQPSO is compared with four exiting improved algorithms in three aspects including mean, standard deviation and best value. Based on history data of 15 stocks from Chinese security market, particle swarm optimization algorithm, QPSO, cuckoo search, bat algorithm and MQPSO are used to solve portfolio optimization models with minimax type risk respectively. Experiments indicate that whether in benchmarking or in simulation application, MQPSO is better than other swarm intelligence algorithms at convergence accuracy and stability.