Short-term photovoltaic power prediction based on modal reconstruction and hybrid deep learning model

Abstract

Short-term photovoltaic (PV) power prediction is significant in improving power grid planning and dispatching capacity. However, the change of PV power has strong randomness and volatility, which will affect the prediction accuracy. A PV power short-term prediction model is proposed in this paper, which combines Pearson correlation coefficient (PCC), ensemble empirical modal decomposition (EEMD), sample entropy (SE), sparrow search algorithm (SSA), and long short-term memory (LSTM). Firstly, the anomalous data of the PV power plant is cleared and complemented, and the key meteorological features are selected as input using the PCC to realize the dimensionality reduction of the original data. Secondly, the input and output variables are decomposed into components of different frequencies using EEMD. Calculate the SE value of each component, and merge and reconstruct the components with similar SE values. Finally, the LSTM prediction model of each reconstruction component is established. The SSA is used to optimize the LSTM structure parameters, and the optimal parameter combination is selected to reduce the prediction error. The predicted values of the reconstructed component are summed, and the final prediction results are analyzed according to R2, MAE, and RMSE. The results show that the PCC-EEMD-SSA-LSTM model proposed in this paper has a minimum prediction error of PV power under different weather, verifying that the proposed hybrid model has superior prediction accuracy.

Keywords

• Short-term photovoltaic power prediction
• Pearson correlation coefficient
• Ensemble empirical mode decomposition
• Sample entropy
• Sparrow search algorithm
• Long short-term memory