Modeling dynamics from only output data.
We examine the problem of reconstructing input-output systems from time series data. Although the method of delays has already been used in the case where both input and output are measured, in some cases, the inputs cannot be measured, and hence, the method of delays cannot be used. On the basis of ideas derived from existing embedding theorems, we propose to build models by using delays of multivariate observations of output data. Assuming that the inputs are few, we use several observations for obtaining information about the inputs, and the remaining observations for obtaining information about the state of the system. Numerical examples on a discrete map and a continuous-time system show that input-output systems can indeed be identified by using multivariate observations of output data only. We also discuss the application of this method to the analysis of coupled systems or complex networks, by partitioning such large systems and analyzing each subsystem separately. The models used in this paper are nonpredictive models; thus, they cannot be used to predict the future behavior of the system. However, since they model the dynamics of the system, they have other possible applications such as change detection and noise reduction.