Wind energy has its own nature of stochasticity which remains a challenge for wind farm designers to extract energy from it as expected. This happens because the sources of uncertainty are not considered during the design calculations for the sake of simplicity. Such adaptation of a deterministic approach for a stochastic problem often gives erroneous as well unrealistic results, which is not implementable as well as useful during a realistic situation. One approach to handle such uncertainty is to consider the layout optimization as a stochastic optimization problem, which has been solved in this case using robust optimization. As several uncertain realizations are necessary to generate such a robust solution, daily average WFMs obtained from the real-life wind data are collected, and Convolutional Auto-Encoders are utilized to extract latent vectors from them. Using a novel clustering algorithm, these latent vectors are clustered into different groups to generate new WFMs, which in turn can be used for wind farm micrositing under uncertain circumstances.