A short-term wind energy hybrid optimal prediction system with denoising and novel error correction technique

Abstract

As an important part of system operation and wind power station planning, wind energy forecasting is significantly affected by the strong volatility, intermittency and variability of the wind speed sequence itself. Therefore, to improve forecast accuracy and stability, the complex new energy forecasting problem is analyzed, and a research plan including denoising processing, input data feature optimization, model optimization selection and error correction is proposed. Apply the wavelet soft thresholding algorithm (WSTD) to sequence denoising, remove redundant information, and determine the best predictive model and model input for the sequence based on a new model selection principle (MSP) and phase space reconstruction technique (PSR), simultaneously, introduce the bayesian optimization (BO) to design the optimal reference search strategy to improve the learning efficiency of the model. Then, in order to avoid losing prediction information, a decomposition-segment error correction (DSEC) technique is proposed to enhance the veracity and adaptability of wind speed prediction. Based on 3 different data, the improved percentage of mean absolute percentage error is 7.366%, 7.514% and 10.649% respectively compared with the best performing comparative model, which verifies the effectiveness and applicability of the proposed framework and can provide strong support for smart grid planning.