Sizing residential photovoltaic systems in the State of Georgia

This paper proposes a novel photovoltaic (PV) distributed generation design process to optimally size and select the PV system characteristics including energy storage capacities using Artificial Neural Network (ANN). This process is designed to be functional for a wide range of electrical loads and solar irradiance values for residential houses in the State of Georgia. Under this system, two neural network models have been implemented. The first ANN model is used to size the PV house system using inputs such as load requirement (kWh/day) and solar radiation (kWh/m2/day). The outputs of this model are the area needed for PV installation, the peak PV power capacity, number of modules, battery storage capacity, and the battery Ampere Hour. The second ANN model uses the rated power from the first model to select the PV system parameters using a large database of commercially available PV modules. The evaluated PV parameters are summarized as follows: the open-circuit voltage, short-circuit current, maximum voltage and current, cell efficiency, module efficiency, and the number of cells needed for the system. Simulink models were created using a set of algebraic equations that were derived to generate the sizing parameters without the need for retraining the network every time. ANN models were implemented with optimal number of layers and neurons, which were trained, simulated, and verified with 99.99% regression accuracy.