| System identification creates mathematical models of systems using measurements of their inputs and outputs. System identification techniques have been used in a variety of applications including in biomedical, control, and aeronautical engineering. Linear time-invariant models have been widely used in all these applications since they provide a simple description that nonetheless predicts the system's behaviour within acceptable errors. In some cases, linear, time-varying models can be used to provide more accurate descriptions of various non-linear or time-varying phenomena, while still retaining a relatively simple description. This thesis reviews two existing approaches for identifying time-varying systems: Temporal basis expansions and Ensemble techniques. This thesis presents two new algorithms for identifying linear time-varying systems including a novel term-selection technique. The algorithms have been demonstrated by applying them to ankle compliance dynamics and to experimental human elbow stiffness data. |
