The Information Structure of Feedforward/Preview Control Using Forecast Data

Preview control using a fedforward imperfect forecast measurement of a disturbance signal is investigated in the context of discrete-time linear quadratic Gaussian (LQG) control. A new approach for incorporating such forecast measurements is built directly on established preview control models and results. The calculation of the optimal control gain, for which an efficient computation has already been derived, is found to be independent of the stochastic forecast measurements, implying that the optimal state estimator is where performance improvements in this problem set-up occur. Most significantly, the forecast data model is shown to equip the problem with a nested information structure whereby any forecast feedforward control problem of a fixed horizon length is always equivalent to a problem with a longer horizon and infinitely unreliable forecast measurements beyond the smaller horizon length. A numerical example illustrates the effect of forecast horizon length and data quality on the closed-loop system performance.