ABSTRACT

The previous chapter discussed the problem of estimating a transfer function model from noisy step response data. Use of individual step tests to identify models for multivariable systems and systems with nonstationary disturbances is not always practical and may not lead to meaningjul results. In these situations, input signals with multiple and perhaps random moves are required. In the next three chapters, new system identification tools are developed and used to build models for dynamic processes. This chapter discusses least squares parameter estimation using the orthogonal decomposition algorithm and proposes a simplified computational procedure for calculating the PRESS statistic using this algorithm. The PRESS is used extensively in this chapter and later chapters for structure selection of multivariable process models and disturbance models.