Short-term PV Power Prediction Based on Bi-directional Gated Recurrent Unit Network and Adaptive Chirp Mode Decomposition

Abstract

With the increasing proportion of installed PV capacity and access to the system, accurate short-term PV power forecasting is critical to maintain safe and stable operation of grids containing a high proportion of PV power sources. In addition, a single prediction model has limited ability to improve the accuracy. To address the above issues, this paper proposes a short-term PV power prediction method based on bi-directional gated recursive unit (Bi-GRU) network and adaptive chirp modal decomposition (ACMD). First, to reduce the complexity of the historical PV power time series data, the ACMD algorithm is used for signal decomposition. Then a Bi-GRU network is built for the prediction of the sequences after noise reduction processing. The model is validated using actual power output data from a PV plant in Xinjiang, China. By comparing with SVM, ELM and BiGRU prediction models, it is verified that the ACMD-BiGRU model proposed in this paper has better results and can effectively improve the short-term PV power prediction accuracy.