The traditional manufacturing industry is gradually transitioning toward intelligent and environmentally friendly production modes. To achieve efficiency improvements and emissions reduction in flexible manufacturing workshops, this study aims to minimize makespan and total energy consumption. It constructs an integrated scheduling model for a variable-speed AGV and machine under charging constraint. An improved NSGA-II optimization algorithm is designed based on a hybrid learning strategy. This algorithm adopts a four-segment chromosome encoding scheme based on process, machines, AGV and AGV speed, with different crossover and mutation operators for each encoding segment. Additionally, an elite preservation strategy based on opposition-based learning is employed to enhance the algorithm's population diversity. Furthermore, a neighborhood search operator tailored to problem characteristics is proposed, utilizing the Q-learning reinforcement learning algorithm to dynamically adjust the neighborhood structure during the iteration process, thereby enhancing the algorithm's local search capabilities. Finally, the effectiveness of the improved NSGA-II in solving this problem is verified through simulation tests.