To effectively suppress hunting of control output by performing feedback control with a suitable feedback gain meeting the current state of each control object.
An FB computation section 63 computes FB gains (optimal feedback gains) F(1)-F(9) using a predetermined evaluation function for a regulator constructed based on model parameters of a control object model. Based on a control deviation e(i), an FB term ufb(i) is computed according to state feedback control rules using the FB gains F(1)-F(9). A weight coefficient R used for the evaluation function and determining weight of a variation of the FB term ufb(i) is set by a weight coefficient setting section 77. When hunting of control output y(i) is detected, the weight coefficient setting section 77 enlarges the currently set weight coefficient R to reduce the FB gains, thereby suppressing hunting.