ABSTRACT
In Part II, we saw that the innovation approach has been useful for the identi–cation and estimation of stochastic or deterministic dynamical system models from observed time series data, where the use of innovation-based LSE and MLE methods is mathematically supported by a Markov process theory, summarized in the theorem, that says that for any continuous Markov process xt, the prediction error, wt = xt − E[xt|xt−Δt, xt−2Δt, …], converges to a Gaussian white noise for Δt → 0.
