Neural Networks Based Solar Power Forecast in Multi-Area System

At the higher penetration levels, more work is needed to encourage the use of grid-connected photovoltaic (PV) systems as a key source in electric power networks. Based on the idea that using solar power forecast data might enhance the management and performance of small electric networks, this thesis tackles the variability of PV electric generation. An architecture system utilizing neural networks is employed for the Nonlinear Autoregressive shallow network. To evaluate how input configurations, affect the model’s predicting performance, they are compared and optimized. Forecasts up to several hours ahead of time are tested to confirm the model’s forecasting accuracy for various short-term time horizons and compared with the persistence model once the input configuration with the best network performance is chosen. When it came to Mean Square Error (MSE) number, the suggested model fared better than the real model.