A New Approach to Forecasting

<p><span style=background-color: rgba(255 255 255 1); color: rgba(33 37 41 1)>The greatest original work on forecasting ever published. By a master of the post-Kalman era. Professor O'Reilly brings a lifetime's engineering experience and not a little scholarship to an enduring problem. The result: a completely new theory of filtering and prediction for causal dynamical system models subject to significant disturbance uncertainty. Any causal dynamical system model can be used.</span></p><p><span style=background-color: rgba(255 255 255 1); color: rgba(33 37 41 1)>No</span><em style=background-color: rgba(255 255 255 1); color: rgba(33 37 41 1)> a priori</em><span style=background-color: rgba(255 255 255 1); color: rgba(33 37 41 1)> knowledge of the model uncertainties is required. Estimation of uncertain dynamical systems it turns out is a modelling problem. With necessary model validation. The criterion for high-fidelity signal reconstruction is how closely the signal estimates resemble the measured output data of the actual dynamical system.</span></p><p><span style=background-color: rgba(255 255 255 1); color: rgba(33 37 41 1)>In contradistinction to the Kalman off-line nominal design approach the causal estimation approach is an on-line model tuning approach. This physical approach places estimation of dynamical systems on an experimental footing akin to classical physics and engineering. And closer to present day industrial practice. Both causal and Kalman approaches are evaluated within twentieth century filtering and prediction theory. The new estimator is completely general non-statistical and very easy to use.</span></p>