Kullback–Leibler Control in Boolean Control Networks

This article addresses the Kullback–Leibler (KL) control problem in Boolean control networks. In the considered problem, an extended stage cost function depending on the control inputs is introduced; in contrast to a stage cost of the conventional KL control problems in the Markov decision process cannot take into consideration the control inputs. An associated Bellman equation and a matrix-based iteration algorithm are presented. The theoretical analysis shows that the proposed KL control results in an approximated form of conventional dynamic programming (DP). Furthermore, the convergence analysis is presented, with the weight parameter converging to zero and diverging to infinity. In practical application examples, a comparison of the conventional DP and proposed KL control is illustrated.