Prediction Intervals for Short-Term Photovoltaic Generation Forecasts

Because of the volatility and intermittent of photovoltaic (PV) power, in order to meet the requirement of grid planning, the prediction of PV systems not only need to provide the exact outcome of the predicted value, but also need to make a reasonable assessment for the risk including predicted value. This paper proposes a nonparametric method for construction of reliable Prediction Intervals (PIs) based on radial basis function (RBF) neural network forecasts. A lower upper bound estimation (LUBE) method is adapted for construction of PIs. By analyzing the factors of PV power generation, based on similar day principles, the history of power data were selected. Then, a strong association in favor of historical data as a sample model is conducive to convergence. Based on the actual data of test results show that, compared with traditional prediction methods, the proposed uncertainties prediction LUBE method based on RBF network can effectively describes the short-term variation characteristics of photovoltaic power.