The mixed logical dynamical(MLD) framework introduces a new way to handle infeasibility and constraint prioritization issues in predictive control. But the optimal algorithm is characterized with computation complexity and long resovling time. In this paper, a penalty function method is introduced in the index function, and then the mixed integer quadratic programming(MIQP) problem is converted into quadratic programming(QP) problem. Thus an improved optimal algorithm is developed. The resovling time is reduced greatly, and computation efficiency is achieved. The proposed algorithm is applied into the Wood-Berry column and promising results are achieved.