Abstract: In the acetic acid concentration, the temperature cannot be well controlled by indirect control. By using wavelet kernel extreme learning machine, this paper proposes a model predictive control strategy (KMPC), which directly control the quality of products on the basis of acetic acid soft sensor. It is known that wavelet kernel extreme learning machine (KELM) has the characteristics, e.g., fast training speed, high precision, and strong generalization ability, which is utilized in the proposed KMPC to establish the prediction model of the acetic acid concentration. Moreover, the predictive controller is used as the master controller for the acetic acid concentration and a PID controller is used as the slave controller for the reboiler vapor flow, both of which constitute a cascade control system. Besides, the sensitive temperature, bottom temperature, inlet temperature, and pressure of reboiler are taken as the disturbance variables to achieve the feedforward control and nonlinear predictive control to the complex distillation column. By means of the simulation based on ASPEN DYNAMICS simulation software, it is shown that, compared with the traditional dynamic matrix control (DMC) method, the proposed strategy can improve the control efficiency of the acetic acid concentration and is easily applied due to its simple structure.