Combination of data sets for system identification
Approaches to system identification can involve the estimation of parameters in a number of separate single-input models or the estimation of a single model using multiple-run techniques in which several data sets are combined. Multiple-run identification can be achieved through application of the superposition principle but this approach is restricted to linear systems and requires test records which are all of the same length. Here a multiple-cost approach is proposed which is found to be extremely powerful and useful. A relationship between the results obtained using this approach and those obtained using straightforward individual-run identification is derived, and its accuracy investigated. Using this relationship, it is possible to obtain estimates for multiple-cost identification results on the basis of individual-run results, allowing existing single-run software and techniques to be used unchanged for multiple-cost identification. An example is given of the use of multiple-cost identification with flight data for a Puma helicopter; good results are obtained and some important issues highlighted and discussed.
