Short-Term Photovoltaic Power Prediction Based on VMD-mRMR and TCN-BIGRU-ATTENTION

Accurate prediction of photovoltaic power generation is of great significance to its reasonable consumption in response to the problems of noise interference and unstable prediction effects of single models in PV time series prediction, this paper proposes a combined prediction model based on TCN-BIGRU. Firstly, an unsupervised clustering algorithm K-means is used to cluster historical data into three typical weather conditions: sunny, partly cloudy, and rainy. The test set is divided according to the clustering results for model training. VMD is utilized to decompose the environmental factor sequence into various characteristic fluctuation sequences, and the max-relevance and min-redundancy (mRMR) method is employed to select the key influencing factors affecting PV power generation, thereby reducing the dimensionality of model input parameters. The selected significant feature subsets are input into the constructed TCN-BIGRU model, and an attention mechanism is added to enable the model to automatically learn key time periods or features, thereby improving prediction performance. Experimental results based on real data demonstrate that the proposed combined model achieves higher prediction accuracy under different weather conditions.