On Time Series Forecasting Error Measures for Finite Horizon Control

Time series forecasting is routinely utilized to improve regulation in finite horizon control (FHC) problems by forecasting the system's uncontrollable inputs. In this brief, we propose a novel measure for validating forecasting models for FHC applications. Specifically, for the case of linear-quadratic time-invariant systems, we derive a closed-form equation for the increase in cost due to forecast error, present techniques for reducing its computational cost, and demonstrate that compared with conventional error measures, model validation using this measure can improve the controller's performance.