Multistep Forecasting for Short-Term Wind Speed Using an Optimized Extreme Learning Machine Network with Decomposition-Based Signal Filtering

Since wind fluctuates with strong variation even within a short-term period, it is quite difficult to describe the characteristics of wind or to estimate the power output that will be injected into the grid. In particular, short-term wind speed forecasting, an essential support for regulatory actions and short-term load dispatching planning during the operation of wind farms, is currently regarded as one of the most difficult problems to solve. This paper contributes to multistep forecasting for short-term wind speed by developing a three-stage hybrid approach named MECE; it is a combination of the ensemble empirical model decomposition (EEMD) method, cuckoo search (CS) algorithm, and extreme learning machine (ELM) method. As the first stage of the hybrid MECE approach, a signal filtering based on a decomposition and reconstruction strategy is adopted and copied by the EEMD method, and a denoised series can be obtained. Then, the CS-optimized ELM is designed as a novel learning method to construct a single layer feed-forward neural network (SLFN); the input weights and biases are determined by the CS algorithm instead of the random initialization within the original ELM. Next, a training and forecasting stage is taken; three different strategies are adopted for multistep forecasting. The chosen data sets are half-hour wind speed observations, including 16 samples, and the simulation indicates that the proposed MECE approach performs much better than the traditional ones when addressing short-term wind speed forecasting problems.