Forecasting Error Modelling for Improving PV Generation Prediction

Abstract

Accurate forecasting of Photovoltaic (PV) generation output is important in operation of high PV-penetrated power systems. In this paper, an adaptive uncertainty modelling method for forecasting error is proposed to improve the prediction accuracy of PV generation. The proposed method models the uncertainty in forecast data using Kernel Density Estimator and guarantee the provision of accurate expected value. Neural Network model is then constructed by the developed uncertainty model to forecast the PV output. The actual confidence level is traced within the day and injected as an input to the Neural Network model by observing the Mean Absolute Prediction Error (MAPE) and Unscaled Mean Bounded Relative Absolute Error (UMBRAE). The proposed method is tested with various significant changes of weather condition and proved to have promising performance on PV generation forecasting. Thus, the developed adaptive uncertainty model can be further used in power system planning that have high-penetration energy sources with stochastic behavior.