An Analysis on Time Intervals and Forecast Horizons for Short-Term Solar PV Forecasting

Chengjie Xu, Nattawat Sodsong, Kunming Yu, Ouyang Wen


The main challenge in solar photovoltaic (PV) generation system is the difficulty in managing power output caused by the rapid fluctuation of solar irradiance. Predicting solar PV power is a potential solution for solar PV operators, which helps improve the overall performance of solar plants. In this paper, different time intervals of data are used as the input in the learning algorithms to determine the optimized time interval. The algorithms used for prediction were Gated Recurrent Unit (GRU) networks, Feed-forward Artificial Neural Network (ANN), Support Vector Regression (SVR), K Nearest Neighbors (KNN), and a hybrid model with the combination of a cascade model and GRU. Each algorithm was compared with each other at different time interval and forecast horizons with root mean squared error (RMSE). The results showed that single step perdition has higher accuracy than multiple steps prediction. In addition, the data time interval has a slight impact on the result but the amount of data has significantly more impact on the outcome. When forecasting for one hour, the hybrid model had the best prediction accuracy with 9.42%.


Full Text:



  • There are currently no refbacks.