Wind power generation is considered an important green renewable energy source due to its ability to meet the world's power requirements over time. However, optimizing the layout of a wind farm to alleviate the wake effect and maximize power generation is a challenging optimization problem due to being non-convex and NP-hard. Several optimization approaches have been recently proposed for tackling this problem; however, they still suffer from low quality of final results and slow convergence speed. Therefore, this study proposes a new, effective approach, namely GBOT, based on the gradient-based optimizer and the recently proposed encoding mechanism. Despite that, GBOT still suffers from a slow convergence rate as the number of wind turbines increases. Therefore, it is improved by replacing the updating scheme of GBO with a novel one to aid in improving the exploration and exploitation performance along the optimization process. This improved variant is known as IGBOT. Both GBOT and IGBOT are compared with several state-of-the-art methods based on two wind scenarios. This comparison is conducted in terms of several performance metrics, including best power output, average power output, worst power output, standard deviation, Wilcoxon rank sum test, Friedman mean rank, multiple comparison test, and convergence curve.