A Simple and Reliable Photovoltaic Forecast for Reliable Power System Operation Control

Recently various forecasting methods for photovoltaic (PV) generation have been proposed in the literature. However, these standard methods cannot be successfully and widely used in general due to the fact that they require access to specialized data that are not always and everywhere readily available in practice. Furthermore, prediction accuracy of such methods tends to deteriorate specially due to data scarcity. This paper proposes a simple and reliable PV forecasting method using machine learning and neural networks. Confidence interval (CI) results are specifically provided for the local supply-demand control as well as for the robust power system security. The proposed method uses only weather forecasting data that are provided by the Japan Meteorological Agency (JMA) and which is available to the public. The proposed method maintains a high level of accuracy by using real-time correlation data between the specific target and the neighboring areas. Multiple neural networks are constructed based on a weather clustering technique. It has been confirmed through extensive simulation results that the proposed method demonstrates robustness in prediction accuracy and CI effectiveness. (C) 2018, IFAC (International Federation of Automatic Control) Hosting by Elsevier Ltd. All rights reserved.