Long-term microgrid expansion planning with resilience and environmental benefits using deep reinforcement learning

Abstract

Microgrid plays an increasingly important role to enhance power resilience and environmental protection regarding greenhouse gas emission reduction through the widespread applications of distributed and renewable energy. Because of the steady growth of load demand, the strict power resilience requirements and the pressing need of carbon emission reduction, microgrid expansion planning considering those factors has become a currently topical topic. In this study, a new framework for long-term microgrid expansion planning, in which a microgrid serves as a backup power system in the event of main grid outages from the perspectives of economy, resilience and greenhouse gas emission, is proposed. Deep reinforcement learning method is used to solve this dynamic and stochastic optimization problem by taking into account various uncertainties and constraints for the long-range planning. Case studies of 20-year microgrid expansion planning using actual data are conducted. The simulation results demonstrate the effectiveness of the proposed framework on reducing greenhouse gas emissions and total cost including economic losses resulting from power grid outages, investment and operating cost of microgrid entities. In addition, the impact of customer load demand and microgrid entities price on optimal planning policies is discussed. The results demonstrate that microgrid expansion planning can be effectively adapted to different levels of load demand and different scenarios of price changes under the proposed framework. This work is helpful for decision makers to implement cost-effective and power resilient microgrid expansion planning with greenhouse gas emission reduction benefits in the long term.