In order to address the stagnation problem in the late stage of evolution of the standard sine cosine algorithm (SCA), this paper makes the analysis of population diversity. The analysis results show that the control factor affects directly population diversity and is decreased exponentially with increase of iterations in the expression of population diversity. In order to improve the ability of exploration and exploitation in the late stage of evolution of the SCA, a multi-scale sine cosine algorithm (MSCA) is presented. In the MSCA, an adaptive multi-scale control factor is designed to regulate population diversity for achieving the search with different layers. Meanwhile, an assisted swarm is developed to coordinate the local search for accelerating the convergence speed and improving calculation accuracy. The assisted swarm evolves independently and each individual can learn from the best experience of the mast swarm or the assisted swarm. The MSCA is evaluated on 23 benchmark functions and compared with the improved versions of the SCA and new swarm intelligence algorithms. The numerical results show that the MSCA can better coordinate the exploitation and exploration capabilities and improve the global optimization ability.