PID-type controllers have been widely used in many industrial applications. Regulation properties of those can be improved through the addition of the Bang-Bang action. In spite of the potentials of this PID-plus Bang-Bang controller, their regulation properties are still limited if a fixed window limit is used in selection of a control action between PID and Bang-Bang action. Thus, this paper proposes an approach to improving regulation properties. Our approach changes window limits adaptable to plant dynamics by use of a Gradient Based Prediction Model. We experimented our control scheme with a DC servo-motor system. It has been shown through experiment that our control scheme outperformed than existing one in terms of overshoot, rise time and settling time.