Abstract. Along with the increasing penetration of wind power into power systems, more accurate forecast of wind power becomes more and more important for real-time scheduling and operation. This paper proposes a novel model for short-term wind power forecast based on singular spectrum analysis (SSA) and self-organizing maps (SOM). In order to deal with the impact of high volatility of the original time series, SSA is utilized to extract the mean trend from the original time series. After that, SOM is applied to select the similar segments from mean trend, which are then employed in local prediction by support vector regression (SVR). Simulation studies are conducted on real wind power time series, and the final results indicate that the proposed model is more accurate and stable than other models.
IntroductionIn recent years, wind energy as a non-pollution, promising type of energy has been the fastest growing renewable energy technology [1]. Obviously, with the increasing popularity of wind power inputting into power systems, the intermittency and randomness of wind have brought great challenge to the safe and stable operation of power systems [2]. Therefore, more accurate and stable forecast is required to guarantee the stability and security of power system operation without increasing the operating cost [3].Many research works have been focusing on improving the accuracy of wind power forecast, and several forecast models have been proposed, which can be approximately classified into two categories: physical models and statistical models [4]. Physical models are usually referred to as meteorological prediction of wind power, which is related to the numerical approximation of the models that describe the state of atmosphere [5]. Unlike physical models, statistical models are established by discovering the relationship of historical data, whose advantages for wind power forecast are the simplicity of model construction, reduced computational requirement, and better performance than physical models in short-term forecast [6]. Therefore, in order to provide a more accurate guidance for power system real-time operation, a short-term forecast model belonging to the statistical models is proposed in this paper.In order to gain more accurate forecast results, it is crucial to reduce the impact of strong volatility of wind power. To achieve this, a number of methods have been applied to extract the mean trend from the original time series. Empirical mode decomposition (EMD) [7], which is a typical approach among them, decomposes the original time series into multiple intrinsic mode functions (IMFs) and considers the first IMF or the first several IMFs as noise [8]. However, this practice is based on experience without theoretical guidance. In this paper, SSA, which is a powerful technique for