Neural network-based solar irradiance forecast for peak load management of grid-connected microgrid with photovoltaic distributed generation

Photovoltaic (PV) power generation is greatly impacted by weather conditions. Even the cast of a moving cloud on the PV panels may cause a dip in generated power. Thus PV power generated forecast is a challenging task but is the key importance to power management of renewable distributed generation. In this work, instead of directly predicting PV generation, solar irradiance is forecasted using a neural network-based model. The aims are to utilize the already well-recorded meteorological data to: (1) enable neural network training and (2) confirm accuracy and effectiveness of the proposed method. The forecasted irradiance is later translated into power that could be converted by a PV system using simple algebraic relation. Input variables of the artificial neural network (ANN) algorithm are made up of the aforementioned meteorological data with the addition of direct visual observation to determine the sky clearness index. The importance of this algorithm is the combined used with short term load forecast (STLF) to provide system operator with confident in predicting the near future state of the system for peak load management.