A Novel Data Augmentation and Swift Optimal Sizing Framework for PV-based EV Charging Microgrid

Optimizing the size of photovoltaic (PV) and energy storage system (ESS) is a key issue to ensure reliable, resilient and economic operation of the microgrid (MG). However, multiplicity of electric vehicle (EV) load profiles and computational complexity of the optimizing models limit the application of existing methods upon restricted scenarios. In this paper, a novel framework combining physical-economical model (PEM) and data-driven model (DDM) is established. First, a data augmentation approach based on reversed Monte Carlo (RMC) is proposed to generate comprehensive synthetic EV load datasets for big data training. Subsequently, a PEM is given to determine the capacity configuration of PV and ESS under each load condition, with the objective to obtain the best financial profitability. The model takes into consideration battery degradation and optimal power dispatch. Finally, a DDM is trained with machine learning techniques to relate optimal sizes with input of EV profiles and MG features. Testing results illustrate that in the test sets, the neural network could predict the optimal configuration within 15% and 20% relative error, respectively, enormously narrowing design space of PEM, and thus produces over 88% runtime savings while maintaining high fidelity and optimality.