Load balancing of dynamical distribution networks with flow constraints and unknown in/outflows

TL;DR

This paper develops a distributed control approach for load balancing in dynamical distribution networks with flow constraints and unknown constant inflows/outflows, ensuring convergence to consensus.

Contribution

It introduces a port-Hamiltonian based Lyapunov method for load balancing under flow constraints in directed networks, extending previous models to more realistic flow limitations.

Findings

Distributed PI controllers achieve load balancing despite unknown inflows/outflows.

Necessary and sufficient conditions for load balancing with flow constraints are derived.

The approach guarantees convergence to consensus in constrained flow networks.

Abstract

We consider a basic model of a dynamical distribution network, modeled as a directed graph with storage variables corresponding to every vertex and flow inputs corresponding to every edge, subject to unknown but constant inflows and outflows. As a preparatory result it is shown how a distributed proportional-integral controller structure, associating with every edge of the graph a controller state, will regulate the state variables of the vertices, irrespective of the unknown constant inflows and outflows, in the sense that the storage variables converge to the same value (load balancing or consensus). This will be proved by identifying the closed-loop system as a port-Hamiltonian system, and modifying the Hamiltonian function into a Lyapunov function, dependent on the value of the vector of constant inflows and outflows. In the main part of the paper the same problem will be addressed for the case that the input flow variables are {\it constrained} to take value in an interval. We will derive sufficient and necessary conditions for load balancing, which only depend on the structure of the network in relation with the flow constraints.