Smart algorithms for power prediction in smart EV charging stations

Abstract

Power prediction in solar powered electric vehicle (EV) charging stations is very essential for smooth and uninterrupted operations due to the high oscillatory output of renewables and their dependence on various atmospheric factors. The need for early prediction helps EV stations improve their power performance and utilize available power by designing intelligent charge scheduling algorithms. This study introduces a novel design approach for an off-grid photovoltaic (PV)-powered EV charging station, which involves three main stages: evaluating and analyzing different solar irradiance prediction models (theoretical, empirical, and artificial neural network (ANN) models), forecasting day-ahead solar power profiles, and optimizing charge scheduling for pre-booked vehicles using energy storage systems (ESS). The effectiveness of various solar irradiance prediction models is assessed to identify the best-performing model. The proposed approach employs a novel algorithmic procedure to fine-tune the selected model using a basic dataset. Power prediction simulations are conducted using MATLAB, while Python is utilized for model development. The feed forward neural network (FFNN) model for irradiance prediction has a 0.88 R² score; the anisotropic general regression neural network (AGRNN), isotropic GRNN both have 0.94 and 0.95 R² values for direct PV current prediction, providing a strong base for reliable forecasting models. The significance of ESS backup for effective charging stations is clearly demonstrated by a remarkable 20 kW peak shaving.