Representation learning with extreme learning machines and empirical mode decomposition for wind speed forecasting methods

摘要

Time series analysis has become more accurate with the emergence of powerful modelling methods based on machine learning development. Prediction models use historical time series to predict future conditions that occur over periods of time. However, most of these models are shallow models, only containing a small number of non-linear operations and without the ability or the capacity to extract underlying features from complex time series accurately. Moreover, deep learning approaches outperform statistical and computational approaches if a large amount of data and/or hidden layers are involved in the development of a forecasting model, but they are criticized for their relatively slow learning speeds. Therefore, this research proposes a hybrid model, which is hybridized by empirical mode decomposition, stacked auto-encoders, and extreme learning machines, aiming to forecast wind speed accurately and efficiently. The evaluation is undertaken by conducting extensive experiments using real-world data. The results show that the proposed E-S-ELM can accurately and efficiently forecast wind speed, and the effectiveness of the shared-hidden-layer approach for deep networks is also demonstrated.