Combined ultra-short-term photovoltaic power prediction based on CEEMDAN decomposition and RIME optimized AM-TCN-BiLSTM

Abstract

Photovoltaic power prediction is crucial to the stable operation of the power system. In order to further improve the accuracy of photovoltaic power prediction, a Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) and Rime-ice (RIME) optimization algorithm and optimization Attention Mechanism (AM)-Time Convolutional Network (TCN)-Bidirectional Long Short-Term Memory Neural Network (BiLSTM) combined ultra-short-term photovoltaic power prediction model is proposed. First, the original power sequence is decomposed using CEEMDAN to obtain smoother data; then, for the inherent intermittency, variability, and stochasticity of PV power generation, a combined AM-TCN-BiLSTM prediction model is constructed to extract features and learn the PV power, and the RIME simulates the growth and crossover behaviors of the mistletoe-particle populations with powerful global optimization functions. The hyperparameters of the prediction model are optimized by the RIME algorithm, and the optimized hyperparameter prediction model is used to predict each subsequence obtained from the decomposition. Finally, the prediction results of each sub-sequence are integrated and reconstructed to obtain the final PV power prediction value. The simulation verification shows that the model can effectively improve the prediction accuracy compared with the comparison algorithm. In the experimental results, the MAE for the first and second predictive steps were recorded as 4.3116 and 5.0342, respectively. The RMSE values for these steps were 6.7357 and 8.5834, respectively. Additionally, the R² showed a significant improvement, reaching 0.9879 for the first step and 0.9803 for the second step. These outcomes validate the effectiveness of the model proposed in this paper.