Statistical wind speed forecasting models for small sample datasets: Problems, Improvements, and prospects

Abstract

Wind speed forecasting models have seen significant development and growth in recent years. In particular, hybrid models have been emerging since the last decade. Hybrid models combine two or more techniques from several categories, with each model utilizing its distinct strengths. Mainly, data-driven models that include statistical and machine learning models are deployed in hybrid models for shorter forecasting time horizons (<6hrs). Literature studies show that machine learning models have gained enormous potential owing to their accuracy and robustness. On the other hand, only a handful of studies are available on the performance enhancement of statistical models, even though hybrid models are incomplete without statistical models. To address the knowledge gap, this paper addressed the problems of traditional statistical models while enhancing prediction accuracy. Initially, the multi-step ahead wind speed forecasting performance of eight commonly used statistical approaches is evaluated using four different case studies and three rolling windows. The reasons for erroneous wind speed forecasts are discussed in detail. Next, four enhanced models were considered while addressing the shortcomings of conventional methods. In addition, four machine learning models are also analyzed for comparison. Moreover, the outcomes of the comparisons are discussed, explaining the higher prediction accuracy of improved models. A Global Performance Indicator (GPI) is used to rank the investigated wind speed forecasting models. Results showed that exponential smoothing models have a greater GPI in most cases, whereas Markov Chain models are among the poorest. The grey models are more suitable for smaller samples of data. In machine learning models, Support Vector Machine (SVM) has proven to be the best choice. Overall, the improved models show between 4% and 28% higher accuracy than their counter traditional models. Lastly, the future directions are highlighted that need subsequent research to further improve forecasting performance.