抄録
In a linear quadratic optimal control problem, the optimal control is synthesized by a multiplying current states or optimal state estimates by an optimal feed-back matrix which is obtained by means of off-line computation.
If we take account of information costs (information transmitting cost, information storage cost and information processing cost) necessary to generate the control in addition to the quadratic performance index, these information costs will increase considerably as the size of the state dimension increases.
This paper treats a suboptimal control approach to the optimization problem of a linear discrete time system by use of deterministic patterns instead of true states from a viewpoint of decreasing information costs and presents the degradation of performance index value caused in the suboptimal control. It is shown that this degradation is formulated as the sum of errors in a quadratic form. Each error vectol arises from patternization at each stage of computation.
