Optimized short-term wind power forecasting based on a combination feature selection and greater cane rat algorithm

Abstract

Listen

Translate article

With the swift advancements in renewable energy, wind power as an important clean energy has been paid more and more attention. Due to the intermittency and volatility of wind energy, accurate wind power forecasting has become crucial for improving the stability of grid integration. This paper employs a combination of maximal information coefficient (MIC) and SelectKBest methods for feature selection, along with a greater cane rat algorithm (GCRA) to tune the hyperparameters of long short-term memory (LSTM), establishing the MIC-KBest-GCRA-LSTM model. This model not only overcomes the limitations of traditional feature selection methods, which consider only the correlations among features or the correlations between features and the target variable, but also proposes an optimization method with stronger global search capabilities for optimization problems that simultaneously involve continuous and discrete hyperparameters. To confirm the proposed model’s effectiveness, six control models are set up, and their prediction results are compared. The results show that the proposed MIC-KBest-GCRA-LSTM model reduces the root mean square error (RMSE) by 32.4% compared to the baseline LSTM method. Additionally, the prediction accuracy of this model has significantly improved compared to the other five models.