Structured model reduction

In this paper, we consider model reduction by structured aggregation, where each reduced order state component is constrained to be simply a sum of components of the original state. This requirement, designed to preserve the physical significance of the system variables, is commonly imposed on models of large power systems, chemical process systems, and economic systems. We examine the principle of invariant response as a criterion for partitioning the state into components to be lumped together. It is shown that very coarse system responses can be used to determine appropriate partitions for the structured aggregates.