Power Generation Prediction of Distributed Photovoltaic Systems Based on Machine Learning

The application of photovoltaic power generation technology in the energy field is becoming more and more extensive, In order to better assist in the management and scheduling of the power grid, it is essential to enhance photovoltaic power generation prediction technology. The photovoltaic power prediction based on bidirectional recurrent neural network (BiGRU) model has shown strong performance in recent years. We hope to enhance accuracy by optimizing the model to better apply this model to power generation prediction. In this study, we used K-fold cross-validation to optimize the model training process, and tried to use multi-step recursive prediction instead of conventional single-step prediction as the prediction method. Finally, we also adjusted the hyperparameters of the model, resulting in a reduction of the root mean square error (RMSE) by 12.316%