合意制御に基づく分散事象駆動型劣勾配アルゴリズムによる制約なし凸最適化

抄録

This paper considers a distributed unconstrained optimization problem where each agent has a local convex cost function and the sum of these functions is defined as a global cost function. We propose an event-driven subgradient algorithm based on consensus control to minimize the global cost function. Each agent has an estimate of the optimal solution as a state. In the proposed algorithm, each agent sends its state to the neighbor agents only at trigger-times when the error of its state exceeds a threshold. We show that the error between the estimate of the global cost function of the agents given by the proposed event-driven algorithm and the optimal cost is upper bounded. The simulation results show that the convergence speed by the proposed event-driven algorithm improves and the number of trigger-times can be reduced compared with the existing subgradient methods.