Multi-agent Q(λ) learning for optimal operation management of energy internet

Abstract

This paper proposes an optimal operation management methodology based on the multi-agent reinforcement learning (MARL) in energy internet (EI). An integrated approach to minimize the total cost of operation of such an electrical, natural gas and district heating network simultaneously is studied. A novel multi-agent Q(λ) learning method is presented to form a coordinated optimal management strategy of energy internet with multiple We-Energy(WE), and an equal interval sampling method is proposed to find the optimal discrete action sets so as to enhance the performance of the control areas. Furthermore, a global Q operator is designed to produce a global Q function considering the local reward from each agent which optimizes simultaneously. The proposed method verified by case studies applied to the modified energy network. Compared with the centralized approach, the test results show that the proposed method can provide a fast solution for the optimal operation management which can be applied to multiple We-Energy internet with sufficient accuracy.