Radiotherapy planning is a critical aspect of cancer treatment, where the optimal selection of beam directions and dose distributions significantly impacts treatment efficacy and patient outcomes. Traditionally, this process involves time-consuming manual trial-and-error methods, leading to suboptimal treatment plans. To address this challenge, optimization strategies based on advanced artificial intelligence (AI) techniques have been explored. This paper presents an investigation into the application of AI-driven optimization methods for beam direction and dose distribution selection in radiotherapy planning. The study proposes an approach utilizing Convolutional Neural Networks (CNN) to learn the relationship between patient anatomy and optimal beam orientations. The CNN model is trained on a dataset comprising anatomical features and corresponding beam orientations, derived from a column generation (CG) algorithm. Additionally, Particle Swarm Optimization (PSO) and Grey Wolf Optimization (GWO) algorithms are employed to optimize the CNN's weights and biases to attain the Fluence Map Optimization (FMO) objective function. Experiments are conducted using data from 70 clinical prostate cancer patients. The results demonstrate the effectiveness of the CNN-PSO and CNN-GWO approaches in generating beam orientations that yield treatment plans with dose distributions comparable to those obtained through traditional CG. DVH analysis of the resulting plans for different anatomical structures validates the accuracy and feasibility of the CNN-GWO model in radiotherapy planning. The findings of this study highlight the potential of AI-driven optimization strategies to revolutionize radiotherapy planning by significantly reducing planning time and enhancing treatment plan quality.