RSMD-RF-BGSkip Based PV Generation Prediction Method

With the increasing use of photovoltaic (PV) power generation, power forecasting has become equally important to maintain stable and economic operation of the power system. However, the high frequency component of the PV power time series reduces the accuracy of the model predictions. Therefore, this paper proposes a short-term prediction model for PV power based on Row Secondary Modal Decomposition (RSMD), Random Forest (RF), and BGSkip neural network. Firstly, modal features with different complexity are obtained by RSMD. Secondly, RF is used to select different modal features. Then, to improve the prediction performance of the model, the BGSkip model employs a hybrid neural network to accurately predict the nonlinear part, while the linear part is handled by an autoregressive model. The prediction results of these two parts are integrated through the BGSkip model to output more accurate prediction values. Finally, the historical data of PV power plants in a region of Liaoning is utilized for experimental validation. The experimental results show that $R^{2}$ , $E_{MAPE}$ and $E_{RMSE}$ of the RSMD-RF-BGSkip short-term forecasting model are improved by 7.55%, 0.261% and 16.01% respectively compared with the most advanced models, which has higher forecasting accuracy.