Research on PV power prediction model based on data dimensionality reduction and optimization of CEEMDAN-SSA-GRU algorithm

Photovoltaic (PV) power prediction faces challenges due to generation volatility. This study compares extreme learning machine, support vector machine, and gated recurrent unit (GRU) models, identifying GRU as the optimal base predictor for time-series data. We propose a hybrid CEEMDAN-SSA-GRU model, integrating complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) for signal decomposition, sparrow search algorithm (SSA) for GRU parameter optimization, and a hybrid algorithm combining Pearson correlation coefficient (PCC), Spearman correlation coefficient (SCC), and principal component analysis (PCA), hereafter abbreviated as PCC-SCC-PCA, for dimensionality reduction. Experimental results demonstrate the model's superior accuracy across sunny, cloudy, and rainy conditions, achieving the lowest root mean square error and highest R-square (R2) compared to GRU, SSA-GRU, and CEEMDAN-GRU variants. The PCC-SCC-PCA algorithm further enhances computational efficiency without compromising prediction performance. This framework provides a robust solution for grid-connected PV power forecasting.