Energy Crisis and Global warming highlight the exploitation of renewable energy sources such as wind energy. The biggest challenge lies in the stochastic nature of wind, which makes the wind energy conversion systems inefficient. The important aspect in the design and control of wind farm is the probability mass function which is constructed using the wind speed and direction time series, called Wind Frequency Map. Due to the limited amount of available wind time series data, the construction of Wind Frequency Map is inaccurate and results in unrealistic estimation of power production from a wind farm. Though wind speed and direction forecasts can be used for construction of Wind Frequency Map, it needs long-term forecasts for accurate wind frequency maps which is not reliable. Therefore, the authors, in this paper, have explored data-driven probabilistic technique, called Generative Adversarial Networks (GANs) to capture the distribution of wind scenarios from the limited wind time series data. The ability of GANs to capture the hidden distributions from the available data can be utilized to generate new scenarios from the data which leads to efficient design of Wind Energy Conversion Systems under uncertainty by performing Robust Optimization under Bayesian framework.