This study proposes an intelligent music generation model based on Bayesian optimization, aiming to improve the performance of the music generation model through efficient hyperparameter tuning. Bayesian optimization uses a proxy model to intelligently search the hyperparameter space, thereby avoiding the problem of computational redundancy in traditional optimization methods and improving the quality and efficiency of music generation. Through comparative experiments with methods such as random search, grid search and genetic algorithm, the results show that the model based on Bayesian optimization has significant advantages in generation quality (such as spectral distance, note prediction accuracy, innovation and harmony score) and AUC value of generated-true comparison. In addition, Bayesian optimization significantly reduces the time cost of model training and shows higher optimization efficiency. Overall, this study verifies the potential of Bayesian optimization in intelligent music generation and provides an effective solution for improving the quality and optimization efficiency of generated music.