Novel algorithm improves 'consensus' performance in multi-agent systems

3 years ago 293
network Credit: Pixabay/CC0 Public Domain

Researchers person devised a amended web 'topology' wrong distributed multi-agent systems to amended the velocity astatine which their nodes converge connected statement regarding a azygous information worth needed during computation. The technique, devised by researchers with the Sharif University of Technology successful Tehran, is described successful the September 2021 Issue of the IEEE/CAA Journal of Automatica Sinica.

Within machine science, probe into multi-agent systems has enjoyed a large woody of attraction successful caller years, with uses arsenic varied arsenic , supply-chain management, bitcoin and adjacent usage of swarms of drones by the military. A multi-agent strategy is defined arsenic a decentralized of bundle agents that enactment unneurotic to lick problems. For a scope of problems, it tin beryllium hard oregon adjacent intolerable for a azygous histrion oregon centralized strategy to solve, but solutions tin beryllium recovered by a decentralized strategy of aggregate nodes oregon agents.

One of the large challenges facing the tract involves processing strategies for specified a distributed web to coordinate statement amongst the nodes connected a azygous information worth that is needed during computation. Such statement is described arsenic "consensus." Consensus successful the realm of machine subject is antithetic from statement successful the quality world: it involves converging connected a azygous worth by the nodes successful the network, overmuch similar a radical of friends agreeing connected which edifice to spell to.

A cardinal trouble is processing statement strategies that tin inactive guarantee wide reliability successful distributed multi-agent systems successful the look of faults oregon failures successful 1 oregon much of the nodes of the network. The statement indispensable beryllium fault-tolerant.

However, dissimilar the friends choosing which edifice to devour at, the distributed statement occupation amongst nodes is not easy. In a consensus-seeking process, the agents successful a fixed web effort to hold connected immoderate quantity by communicating what they cognize to their neighboring agents. But however bash they place a faulty node? Is different node identifying the faulty node really the 1 that is faulty? Making matters worse, the greater the fig of nodes, the greater the complexity of achieving consensus.

To woody with the distributed statement problem, assorted algorithmic solutions person been developed specified arsenic taking a 2/3rd bulk of the nodes arsenic to what they judge to beryllium the close information value.

These assorted statement algorithms whitethorn inactive look challenges with respect to their performance. One mightiness beryllium much robust than different successful the lawsuit of node failure. Another mightiness beryllium weaker astatine that but respond amended to outer malicious attacks. The researchers astatine Sharif University of Technology were focused connected improving the velocity astatine which multi-agent strategy nodes converge connected consensus.

In general, the stronger the connections betwixt the nodes successful a network, the much enhanced the convergence rate. However, promoting specified connection betwixt these agents imposes further costs specified arsenic vigor consumption. In immoderate real-world multi-agent systems, the batteries powering the agents person precise debased capableness and cannot conveniently beryllium recharged oregon replaced. As a result, reducing vigor depletion to widen the agents' artillery beingness has emerged arsenic a captious contented successful these networks.

"But alternatively of focusing connected the artillery for a fixed network, we thought we could plan a amended web 'topology," oregon however the web is enactment together, for a fixed battery," says Mohammad Saleh Tavazoei, an electrical technologist and corresponding writer for the paper. He is presently a Full Professor with the Department of Electrical Engineering astatine Sharif University of Technology.

The main vantage of their model for an optimized web topology is that reduces the communications needed betwixt the agents successful the strategy portion allowing a complaint for convergence upon that tin beryllium adjusted up and down, depending connected requirements.

In the future, the researchers anticipation to widen their enactment to balancing convergence complaint and connection requirements to multi-agent systems that person heterogeneous agents successful their architecture.



More information: Mohammad Saeed Sarafraz et al, A Unified Optimization-Based Framework to Adjust Consensus Convergence Rate and Optimize the Network Topology successful Uncertain Multi-Agent Systems, IEEE/CAA Journal of Automatica Sinica (2021). DOI: 10.1109/JAS.2021.1004111

Provided by Chinese Association of Automation

Citation: Novel algorithm improves 'consensus' show successful multi-agent systems (2021, August 31) retrieved 31 August 2021 from https://techxplore.com/news/2021-08-algorithm-consensus-multi-agent.html

This papers is taxable to copyright. Apart from immoderate just dealing for the intent of backstage survey oregon research, no portion whitethorn beryllium reproduced without the written permission. The contented is provided for accusation purposes only.

Read Entire Article