Short-term photovoltaic power generation forecasting based on random forest feature selection and CEEMD: A case study

To mitigate solar curtailment caused by large-scale development of photovoltaic (PV) power generation, accurate forecasting of PV power generation is important. A hybrid forecasting model was constructed that combines random forest (RF), improved grey ideal value approximation (IGIVA), complementary ensemble empirical mode decomposition (CEEMD), the particle swarm optimization algorithm based on dynamic inertia factor (DIFPSO), and backpropagation neural network (BPNN), called RF-CEEMD-DIFPSO-BPNN. PV power generation is affected by many factors. The RF method is used to calculate the importance degree and rank the factors, then eliminate the less important factors. Then, the importance degree calculated by RF is transferred as the weight values to the IGIVA model to screen the similar days of different weather types to improve the data quality of the training sets. Then, the original power sequence is decomposed into intrinsic mode functions (IMFs) at different frequencies and a residual component by CEEMD to weaken the fluctuation of the original sequence. We empirically analyzed a PV power plant to verify the effectiveness of the hybrid model, which proved that the RF-CEEMD-DIFPSO-BPNN is a promising approach in terms of PV power generation forecasting.