Neural network based learning method for estimating power output from forecasted irradiance for solar photovoltaic system

In recent years, the fast development of solar photovoltaic (PV) technology strongly indicates that PV generation will become one of the most attractive renewable sources of energy. However, solar insolation is not constant and the power output of PV system is influenced by insolation and other meteorological conditions such as air temperature, cloud cover, humidity and wind speed. This makes solar electricity generation highly variable and uncertain. Forecasting of solar insolation and solar PV plant power output are needed to reliably operate the grid. In this thesis, a model for PV system output power forecasting is developed based on Back-Propagation and Radial Basis Function neural network learning methods. The data from existing solar plant at the University of Texas at San Antonio campus with a total rating of 150kW is used for developing the model. Compared to the existing state-of-the-art models that use the physics-based approach and the Auto Regressive Integrated Moving Average statistical approach, the forecasting results show the proposed model is flexible for short-term and long-term forecasting time-scales and is accurate as verified using real data and evaluated against standard error metrics, Root Mean Square Error (RMSE) and Mean Absolute Error (MAE).