Learning based short term wind speed forecasting models for smart grid applications: An extensive review and case study

Abstract

This paper provides an extensive review of learning-based short-term forecasting models for smart grid applications. In addition to this, the paper also explores forecasting models including physical, statistical, hybrid, and uncertainty analysis models for wind speed forecasting. The learning-based models are classified into three broad categories, namely classical machine learning, advanced machine learning, and probabilistic learning. In this work, 41 different models are employed to forecast the wind speed. Dataset for this case study is collected from the site of Jodhpur, India. Dataset have 8759 sample with five features i.e., wind speed, pressure, humidity, temperature, and dew point. This forecast also includes the seasonal effects. Model accuracy has been tested considering single and multiple features in the input data. A comparative analysis of the performance of these 41 learning-based models is conducted based on coefficient of regression and error indices. It is observed that the performance of these models varies with the variability in the season. On the basis of the evaluation of these models, future recommendations are also framed out. These recommendations target of energy storage planning, energy market and policymakers, and reliability and reserve sizing direction. These recommendations can be utilized by authorities for effective planning and coordination of power.