Prediction of photovoltaic power generation based on parallel bidirectional long short-term memory networks
Zhi Rao, Zaimin Yang, Jiaming Li, Lifeng Li, Siyang Wan
Abstract
Current photovoltaic power generation prediction methods have a uniform structure and are vulnerable to data quality issues, leading to inadequate accuracy in predicting photovoltaic power generation. This study proposes a prediction of photovoltaic power generation based on parallel bidirectional long short-term memory networks (BiLSTM). The method combines three BiLSTMs and deep neural network (DNN) to design a parallel BiLSTM framework for photovoltaic power generation prediction. The method utilizes three BiLSTMs to extract deep features from the data in parallel respectively, and then the features are merged and inputted into DNN. With simple weighting, the DNN offsets the errors of the three BiLSTMs against each other to get a more accurate result. The experimental results show that the root mean square error (RMSE) of the photovoltaic power generation prediction method proposed in this study is 9.5733, which is 13.7688 smaller than the RMSE of the best effect among the compared methods.