This paper discusses a new approach to representing a finite automaton as a combination of a linear state equation with a smaller set of free binary variables (i. e., input variables) and binary inequalities, in order to reduce the computational time for solving the model predictive control problem of a class of hybrid systems. In particular, this paper is devoted to proving that a system representation derived by our proposed method is minimal in the sense that the number of its binary input variables is minimal among system models over all linear equivalence transformations that preserve the binary property of free (input) variables.