Decentralized approach to intrusion detection in dynamic networks of the Internet of Things basing on multi-agent reinforcement learning and inter-agent communication

The paper proposes a multi-agent reinforcement learning technology for intrusion detection in the Internet of Things. Three models of a multi-agent intrusion detection system have been implemented — a decentralized system, a system with the transmission of forecasts, a system with the transmission of observations. The obtained experimental results have been compared with the open intrusion detection system Suricata. It has been demonstrated that the proposed architectures of multi-agent systems are free from the weaknesses found in the usual solutions