Multi-agent Based Energy Balancing Management Algorithm for Smart Grid System

Malla Dinesh Bahadur, Katsuyoshi Sakamoto, Tomah Sogabe


The smart power system focuses on renewable energy sources, which has the potential to reduce the dependence of residential buildings on electricity systems. However, their integration into existing systems increases instability, supply insecurity. Optimizing output schedules and regular forecasting of electricity demand can improve power system stability. But, constantly changing demand for electric power creates problems in scheduling and forecast. For this, it is essential to prioritize how to exploit the immediate deployment of operating units and storage resources online. These processes include optimization and forecasting processes that can address under the umbrella of a multi-agent learning process. The purpose of the proposed multi-agent algorithm in the centralized controller is to learn the policy of maximizing the performance of each agent by ordering it to perform the average of each step based on each agent's reward. In this way, the multi- agent learns to solve and optimize problems. In this paper, we use a multi-agent DQN algorithm by introducing two global states, which play role to communicate among each individual agent to minimize the electricity peak problem and electricity balance between houses by optimal use of storage utilities.


smart grid; reinforcement leaning; multi-agent

Full Text:



  • There are currently no refbacks.