The inherent stochastic behavior of wind necessitates precise wind power forecasting models for the reliable and widespread integration of wind energy into the electrical grid. This paper outlines the systematic methodology for developing these Feed Forward Neural Network forecasting models, which remain an effective solution despite the success of more complex machine learning models. Leveraging extensive historical meteorological data and feature engineering techniques, a new input feature was derived to anticipate wind power. A grid-search algorithm was then employed to determine the optimal architecture for the forecasting models. Linear regression models served as a benchmark against which the efficacy of the final forecasting models was evaluated. A careful balance between accuracy and computational efficiency guided model selection. This research further demonstrates the often debated quantitative analysis comparing single step and multi-step models, highlighting a preference for the latter due to their comprehensive forecasting capabilities. This research also presents a unique investigation into the computational effects and impact of different training step sizes on accuracy.