Abstract: Dense medium separation process occupies a dominant position in the development of coal preparation industry, but the dense medium separation process often has strong nonlinear characteristics due to the complex process and the large disturbance. In this paper, a multi-model switching adaptive control method is proposed for the density loop of dense medium separation. First, random vector function link network (RVFLN) is employed to estimate the unmodeled dynamics of the system, and linear and nonlinear adaptive controllers are designed. Based on them, the switching mechanism is designed to make the controllers switch according to the raw coal quality, and further improves the control performance and accuracy, thereby improving the economic efficiency of coal preparation plants. Finally, simulation experiments with actual data have been carried out on the density loop of dense medium separation, which shows the effectiveness of the proposed method.