Finite-Time Distributed Average Tracking for Multiagent Optimization With Bounded Inputs

In distributed optimization (DO), the designed algorithms are expected to have a fast convergence rate but less computation cost. Moreover, the boundedness of the control inputs is generally required for practical networking agent systems with actuator limitations. Motivated by these observations, we first revisit the well-known finite-time distributed average tracking (FTDAT) problem where a novel sufficient condition on the control gain and the finite settling time estimation are derived based on the minimum cut of the underlying topology. Then, based on FTDAT, three types of discontinuous dynamics with bounded inputs are designed for solving unconstrained and constrained DO problems, respectively. The first algorithm can successfully find the optimal solution for an unconstrained DO in finite time. For DO problems with a common constraint set or separated equality constraints, two projection-based algorithms are designed and utilized to first make the agents' states achieve consensus in finite time, and then drive the common state to the optimum exponentially. Finally, several case studies and extensive numerical simulations are conducted to testify the designed algorithms.

Automated summary

In distributed optimization, algorithms need to converge quickly and have low computation cost. Boundedness of control inputs is required for practical networking agent systems. Based on the finite-time distributed average tracking (FTDAT) problem, three types of discontinuous dynamics with bounded inputs are designed for solving unconstrained and constrained DO problems. The algorithms successfully find optimal solutions and achieve consensus in finite time.