A Multi-Modal Machine Learning Framework for Wind Power Forecasting

The report of the 2024 National People's Congress and the National People's Congress clearly pointed out that the primary task of high-quality development should be grasped, relating with the new quality productive forces. Meanwhile. wind power, as the new quality productive forces of renewable energy, has significantly played the vital role in the global power supply. So it's important for us to explore the ways and likelihoods about how to accurately predict wind power. Nevertheless, on the one hand,it's numerous features involved in wind power that possibly create the dimensions collapse and waste the unnecessary computational resource. On the other hand, Most of the features which involve with too manyt invalid information are easily to lead to the overfittting model. Thus, we made a analysis on a wind power dateset with 140,000 samples and 76 features. The steps are as follows:①In the first place, it's initial dimensionality reduction that we applied LightGBM to analyse the importance of wind power features. Based on the result, we remove the features with low importance.②Subsequently, it's secondary dimensionality reduction we combined MIC and VIF to construct the SIM matrix so as to measure the linear and nonlinear relationships of each feature.③Consequently, Based on the former dimensionality reduction results, we combined convolution and bidirectional long short-term memory network to build the Conv-BiLSTM model is utilized to forecast the wind power. It shows that the method can effectively achieve a high feature dimensionality reduction rate while achieving more efficient power prediction.