Neural Network Model for Aggregated Photovoltaic Generation Forecasting

This paper presents a forecasting model from 1 to 10 days for the aggregated photovoltaic energy production in Spain. The model uses a convolutional neural network which inputs are meteorological forecasts, historical generation data and the location and installed power of existing plants. The model output is the hourly production of the photovoltaic energy production for the whole system for the following ten days. The results of the model can be used for generation scheduling and system operation on one side and for energy trading in the day-ahead market or in derivative markets on the other side.