Non-Gaussian non-stationary wind pressure forecasting based on the improved empirical wavelet transform

In the field of wind engineering, the field measurement on wind pressure is a vital means of wind resistance researches. However, for the field measurement there are some challenges below. The sensors miss partial data due to their failure or service life. Likewise, some locations are difficult to deploy the required sensors. In this paper, a hybrid prediction model of improved empirical wavelet transform (IEWT), particle swarm optimization (PSO) and least squares support vector machines (LSSVM), is developed for purposes of data recovery and spatial extension of non-Gaussian non-stationary wind pressure (complex signal). In this model, IEWT is first proposed to decompose the signals and get rid of their noise components. Meanwhile, LSSVM is utilized to establish forecasting models of the trend component and main components, where their parameters are optimized by PSO algorithm. Then, the single-point and spatial forecastings are carried out to verify the effectiveness of the proposed model. Furthermore, the empirical mode decomposition (EMD), ensemble EMD (EEMD), and empirical wavelet transform (EWT), are exploited to corroborate the advanced de-noising performance of IEWT. The final results indicate that IEWT can effectively reduce the noise interference and enhance the forecasting precision of complex signal.