Recursive estimation on spatially rearranged data for dynamic order determination and nonlinear component detection

Determination of dynamic order of variables is the first step in system identification. Order determination is in general difficult for nonlinear system identification due to the interaction of system structure (unknown orders) and unknown nonlinearity. If the attenuation of unknown nonlinearity is possible, different system structures could then be fairly compared. Guided by this concept, this work uses a recursive estimation to reduce the effect of the underlying nonlinearity on parameter variation, and proposes a sequential nearest neighbor rearrangement to enhance the reduction. The ldquobestrdquo dynamic order will minimize final prediction error with the consideration of the locality of the model parameters. In addition to determining dynamic orders, the sequential nearest neighbor rearrangement is also extended to detect nonlinear components.