Short-term Solar PV generation forecasting using Bi-GRU neural network

Abstract

This study investigates the deep learning-based rolling window approach in forecasting short-term solar PV generation. This study's primary objective is to develop a model for accurate short-term solar power forecasting at high temporal resolution (20-minute intervals). The solar PV generation data obtained from one of the roof-top solar PV systems at IIT Gandhinagar is used to test and validate the forecasting performance of the developed Bidirectional Gated Recurrent Unit (Bi-GRU) neural network based short-term forecast model. This research study contributes valuable insights into adopting deep learning model for solar PV generation forecast and emphasizes the significance of data window size selection in achieving better forecast accuracy. The findings can benefit the development of an improved forecasting model vital for optimizing distributed renewable energy management.