On Polynomially Solvable Constrained Input Selections for Fixed and Switched Linear Structured Systems
Yuan Zhang,Yuanqing Xia,Shenyu Liu,Zhongqi Sun
DOI: https://doi.org/10.1016/j.automatica.2023.111308
IF: 6.4
2023-01-01
Automatica
Abstract:This paper investigates two related optimal input selection problems for fixed (non-switched) and switched structured (or structural) systems. More precisely, we consider selecting the minimum cost of inputs from a prior set of inputs, and selecting the inputs of the smallest possible cost with a bound on their cardinality, all to ensure system structural controllability. Those problems have attracted much attention recently; unfortunately, they are NP-hard in general. In this paper, it is found that, if the input structure satisfies certain 'regularizations', which are characterized by the proposed restricted total unimodularity notion, those problems can be solvable in polynomial time via linear programming (LP) relaxations. Particularly, the obtained characterizations depend only on the incidence matrix relating the inputs and the source strongly connected components (SCC) of the system structure, irrespective of how the inputs actuate states within the same SCC. They cover all the currently known polynomially solvable cases (such as the dedicated input case), and contain many new cases unexplored in the past, among which the source-SCC separated input (SSSI) constraint is highlighted. Further, these results are extended to switched systems, and a polynomially solvable condition, namely the joint SSSI constraint, is obtained that does not require each of the subsystems to satisfy the SSSI constraint. We achieve these by first formulating those problems as equivalent integer linear programmings (ILPs), and then proving the total unimodularity of the corresponding constraint matrices. We also study solutions obtained via LP-relaxation and LP-rounding in the general case, resulting in some lower and upper bounds. Several examples are given to illustrate the obtained theoretical results.