The engineer or scientist is seldom completely ignorant about the physical processes that he or she is trying to understand. For instance, something is usually known about the measurement devices used for logging data (observations). In the absence of any a priori information, it is very difficult to know how to proceed. In this section, we consider a number of methods for learning about dynamical systems. Each method starts with some assumptions about the physical processes they are trying to learn about.

The best known forecasting methods are based on some form of linear
regression and the use of local linear models. For an introduction to
linear models and time series analysis see **
[Chatfield, 1989]** or
** [Gershenfeld and
Weigend, 1993]) **. The latter provides a quick overview of
autoregressive (AR) and moving average (MA) techniques, explains the
limitations of linear models, and considers the differences between
``short-term'' and ``long-term'' prediction.

- Delay Coordinate Embedding
- Bayesian Networks
- Hidden Markov Models
- Machine Reconstruction
- Neural Networks