Estimation of Photovoltaic Power Generation by Using Deep Learning-based Method

Abstract

It is important to predict the power output of distributed energy resources (DERs) like solar photovoltaic (PV) so as to prevent the power variation impact to power systems. In this paper, the techniques of using weather graphs have been introduced for the estimation of PV power generation. First, traditional Heliosat method is introduced. Secondly, a cloud-type method based on several cloud groups classified by different cloud top altitudes and rainfall intensities is presented and integrates with look-up-table mechanism to determine the PV power generation. Finally, this paper further proposed a deep learning-based method for overcoming the limitations of using above-mentioned methods. In proposed method, not only BILSTM neuron network but also a time mark technique are considered. To validate the performance of proposed method, Experiments based on the PV power generation data collected from a real PV site are included. Analysis results show nRMSE of cloud-type method is 16.83%, which is not better than Heliosat method of nRMSE 6.61%. On the contrary, the nRMSE of 4.67% is obtained from proposed deep learning method that presents the excellent performance among all methods.