Abstract: Traditional intelligent optimization algorithms easily fall into local optimal and premature convergence for cloud computing task scheduling. Aiming at this shortcoming, this paper proposes a reverse learning behavior particle swarm optimization (RLPSO) algorithm. Firstly, a grouping strategy is adopted to divide the individuals into two different groups according to the gender of individuals and the individuals in different groups will perform different search behaviors. This strategy can increase the diversity of search mode and enhance the search ability. Secondly, the reverse learning and mating mechanism are introduced to avoid the local optimum. By means of the reverse learning mechanism, individuals will reversely learn from the target individuals with a certain probability, which can increase the ability of finding a new potential better solution. By using the mating mechanism, the worst individual in the swarm will be replaced by a new better individual so that the global search ability can be further enhanced. Moreover, the convergence of the RLPSO algorithm is discussed in detail. It is proven that the RLPSO algorithm will converge to an equilibrium. Finally, the effectiveness of the RLPSO algorithm is verified by simulation results. It is shown from the experiment results that the RLPSO algorithm exhibits better search performance than the other four algorithms for the large-scale task scheduling.