Abstract: In response to the non-stationary nature and noise interference caused by weather affected photovoltaic power generation,as well as the spatiotemporal coupling effect,a short-term power prediction composite model VMD-CNN-BiLSTM-RF based on Variational Mode Decomposition (VMD),Convolutional Neural Network (CNN),Bi-directional Long Short-Term Memory(BiLSTM),and Random Forest (RF) is proposed.This model uses VMD to decompose the original power sequence into multiple stationary sub-modes,reducing the impact of noise and non-stationarity.It combines CNN with BiLSTM to simultaneously capture spatial features and temporal dependencies of time series data,improving the accuracy of predictions.RF is integrated with sub-model prediction results to enhance the model's generalization ability.An experimental environment was set up on the Matlab platform to conduct comparative experiments and error analysis for the VMD-CNN-BiLSTM-RF combined model.The results show that this model  has significantly improved the accuracy and robustness of short-term photovoltaic power generation prediction.

Key words: photovoltaic power, generation prediction, variational mode decomposition, convolutional neural network, bidirectional long-term and short-term memory network, random forest