Predicting Solar Performance Ratio Based on Encoder-Decoder Neural Network Model

Fault diagnosis of photovoltaic (PV) arrays is an important task for improving the reliability and safety of the overall photovoltaic system. In the traditional fault diagnosis, most of the past researches focus on the I-V curve to determine the PV arrays state. However, most inverters as smart meters cannot obtain data in high frequency and high volume in order to calculate the I-V curve, and the information usually cannot be retrieved conveniently. In this paper, we evaluate various Recurrent Neural Network Models' from the recent advances in the deep learning field and their abilities in predicting PV arrays power generation. The results show the Encoder-Decoder model using Convolutional Neural Network (CNN) as the decoder and Temporal Convolutional Network (TCN) are more effective in predicting the PV arrays solar output production and can be used as the foundation for fault diagnosis and prediction.