Fig. 7. Control-oriented CNM.

(A) CNMc iteratively propagates the state in the phase space populated with the centroids from the two operating conditions with the closest control parameters. (1) Neighboring centroids to the current state ${\widehat{\mathit{x}}}^{\mathit{n}}$ at iteration n are first identified. (2) Their transition properties are calculated and then (3) averaged to determine the next state ${\widehat{\mathit{x}}}^{\mathit{n}+1}$. CNMc accuracy is demonstrated by the autocorrelation function distributions of the data (black) and the predicted case (red) for the (B) Lorenz system and the (C) actuated turbulent boundary layer, respectively.