Improving the Resource Allocation in IoT Systems Based on the Integration of Reinforcement Learning and Rule-Based Approaches in Multi-agent Systems

In recent years, the spread of Multi-agent Systems (MAS) and Reinforcement Learning (RL) as a subdomain of Machine Learning (ML) has affected many fields, such as Cyber-Physical Systems (CPS) and the Internet of Things (IoT). There are many Multi-agent Reinforcement Learning (MARL) applications in CPS and IoT, including energy management, smart transportation systems, job scheduling, etc. Due to the fact that the RL and MARL are slow processes, it is necessary to improve and accelerate them. In this study, we review some applications of MAS in the CPS and IoT. To improve the decision-making in MAS, we apply a combination of learning and reasoning (rule-based). We tested the performance of our proposed method in smart homes for 4 distinct tasks, including Face Detection and Window Blind Control, Finger Touch and Gate Control, Weather Check and Thermostat Control, and Temperature Check and Start Fan to assess its efficacy. We applied our method to the job scheduling problem using two parameters: process unit and bandwidth. The results reveal that the proposed approach performed better than the bankruptcy methods based on these criteria.