Abstract
On regional scale, the estimation of the solar power generation from the environmental conditions and the solar power forecast is essential for Distribution System Operators, Transmission System Operator, energy traders, and Aggregators. Indeed, high photovoltaic penetration results in a stochastic variability of the electric demand that could compromise the stability of the grid, increase the amount of energy reserve and the energy imbalance cost.
In this context, two upscaling methods and several models were developed and used for estimation and forecast of the photovoltaic distributed generation in a small area of Italy with high photovoltaic penetration. The upscaling methods are based on chains of data-driven models (k-mean clustering, principal component analysis and ensemble of neural networks) that use as inputs satellite derived irradiance and numerical weather prediction. These models allow a complete assessment of the accuracy of the regional power forecast at different horizons, from 0 hour ahead (power estimation) to 48 hours ahead. The power estimation model provided a RMSE of 3% of installed capacity. Intra-day forecast (from 1 to 4 hours) obtained a RMSE of 5% - 7% and a skill score with respect the clear sky persistence from -8% to 33.6%. The one and two days ahead forecast achieved a RMSE of 7% and 7.5% and a skill score of 39.2% and 45.7%. The ensemble smoothing effect on cluster scale was also studied. It was proved that the RMSE reduction achieved on cluster scale due to the ensemble smoothing follows the same exponential law found in [2] for single systems. In this case the smoothing effect reduces the RMSE of day-ahead forecast of 12% with respect to the mean single cluster value. It was also shown that the skill score increases linearly with the size of the region with a rate of 0.1% for km.
Furthermore, a method to estimate the forecast error was also developed. It was based on an ensemble neural network model coupled with a probabilistic correction. It can provide a highly reliable computation of the prediction intervals.