Ensemble stochastic configuration networks modeling method for short-term wind speed prediction and noise reduction

Considering complex problems such as extreme weather and equipment failure, the wind speed information collected by the sensor are often contaminated some noise information, which directly affects the modeling accuracy of the wind power prediction problem. In order to solve such problems, this paper uses the noise reduction method of Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise-Improved Wavelet Threshold Denoising (ICEEMDAN-IWTD) to remove interference information and uses optimized ensemble Stochastic Configuration Networks (ESCNs) to predict wind speed to improve the prediction accuracy. Firstly, a multi-strategy improved coati optimization algorithm (ICOA) is proposed, and then it is applied to the parameter optimization of ICEEMDAN, which could better decompose the noisy information into different time-frequency scales which conducive to distinguish the noise components. Secondly, in order to ensure the elimination effect of redundant information, the wavelet threshold method is used for noise reduction reconstruction, and a new threshold method is proposed to avoid the deviation caused by the traditional threshold function. Finally, the ESCNs method is used to predict the wind speed series, and the weight coefficients of the integrated model are optimized by ICOA to ensure the prediction accuracy of the model, so as to ensure the use efficiency of wind power. In this paper, two real wind speed detection data in Chicago are used to verify the proposed method. The proposed method has better performance in wind power data noise reduction and wind speed prediction accuracy.