Streaming-data selection for Gaussian-process modelling

Abstract

The Gaussian-process (GP) model is an example of a probabilistic, non-parametric model with uncertainty predictions. It can be used for the modelling of complex, non-linear systems and also for the identification of dynamic systems. The output of the GP model is a normal distribution, expressed in terms of the mean and the variance. One of the noticeable drawbacks of a system identification with GP models is the computation time necessary for the modelling. The modelling procedure involves the inverse of the covariance matrix, which has the dimension as large as the length of the input samples vector. The computation time for this inverse, regardless of the use of an efficient algorithm, is increasing with the third power of the number of input data. In this chapter we propose a method for the sequential selection of streaming data so that the size of the active set remains constrained. Furthermore, for better adjustment of the model to the system the hyperparameter values are optimised as well. The viability of the proposed method is tested on data obtained from two, nonlinear, dynamic systems. © Springer-Verlag 2013.