Ultra-short-term wind speed prediction based on empirical wavelet transform and combined model

Due to the variability, proneness to outside influences, and constantly fluctuating nature of ultra-short-term wind speed sequences, wind speed forecasts in wind farms are not precise. A method to predict ultra-short-term wind speed based on the empirical wavelet transform and a combined model is proposed. Correct ultra-short-term wind speed estimations are crucial for preserving the stability of the electrical power system. Firstly, the empirical wavelet transform technique is used to decompose the ultra-short-term wind speed time sequences into multiple modal constituents; secondly, sample entropy can be utilized for calculating the complexities of the decomposed data elements, and those that can be rebuilt with a similar level of sophistication are rebuilt; after this, the sophistication of the rebuilt components is analyzed, and the improved sparrow search algorithm optimized extreme learning machine and gated recurrent neural network are used as prediction models for various sections; then, an improved sparrow search algorithm to optimize the weight coefficients of each prediction model. Finally, the final prediction results are obtained by multiplying the prediction results of each method by the resulting optimal weight coefficients cumulatively. The research objects are three sets of measured ultra-short-term data on wind speeds with analysis intervals of 15 min and 5 min. Following experimental validation, the suggested method outperforms previous single and combination prediction methods in terms of accuracy when making predictions.