Real-Time Microgrid Dispatching Considering Renewable Uncertainties: An Improved Approximate Dynamic Programming Method

The volatility of distributed photovoltaic (PV) and wind turbine (WT) brings great challenge to the real-time dispatching of microgrid. This work aims at solving the problem via an improved approximate dynamic programming (ADP) method. Firstly, a two-stage microgrid dispatching framework is formulated to tackle uncertainty of PV and WT generation with an ADP model which is trained off-line and utilized in real-time dispatching. Secondly, an ambiguity set is proposed to utilize distribution knowledge of renewable generations for the generation of off-line training scenarios. Thirdly, an alternating direction successive projective approximation routine is proposed for the off-line training of ADP model to reduce the impact of initial cost-to-go value function and improve the accuracy of ADP model. Finally, case studies are conducted on the IEEE 37-bus and 123-bus systems to illustrate the effectiveness of the proposed method.